Phylogenetic Procedures

The study of phylogenetic relationships of the Cynodontinae was based on examination of morphological features. Osteological characters are defined and described under Character Description and Analysis and summarized in appendix 1. Osteological preparations were cleared and counterstained for cartilage and bone using a modification of the method outlined by Taylor and Van Dyke (1985). Supplemental sources of osteological data include previously cleared-and-stained specimens, (with some stained solely with alizarin), and dry skeletons.

Species listed below followed by an asterisk (*) provided the morphological basis for estimating phylogenetic relationships within the Cynodontinae, and were also used as the basis for illustrations or specific observations noted in the text. Justification for choice of these taxa is provided below. The remaining species listed provided additional observations on characiform osteological characters. Whenever a character is mentioned in the text for an outgroup genus without a species cited, it refers to species listed below, and does not imply that the character is present in all species of the genus.

The species name is followed by institutional catalog number, the number and standard length (SL) of specimens. Head length (HL) is listed for some skeletal preparations. Specimens are cleared and counterstained unless indicated as being dry skeletons (S):

Cynodontinae:

Anostomidae:

Characidae:

Chilodontidae:

Ctenoluciidae:

Distichodontidae:

Erythrinidae:

Gasteropelecidae:

Hepsetidae:

Lebiasinidae:

Prochilodontidae:

Hemiodontidae:

Hypotheses of relationships were proposed using the cladistic or phylogenetic method first formalized by Hennig (1950, 1966). Detailed explanations about cladistic principles and their operational aspects are available from many sources (Eldredge and Cracraft, 1980; Nelson and Platnick, 1981; Wiley, 1981; Wiley et al., 1991; Swofford et al., 1996).

Parsimony analysis was employed to generate hypotheses of phylogenetic relationships and of character state transformations using the PAUP computer program, version 3.1.1 by D. L. Swofford (1993), and Hennig86 by Farris (1988) associated with Tree Gardener, version 2.2 (Ramos, 1997). The small number of taxa in the analysis permitted the use of the exhaustive search option, which evaluates every possible combination of the taxa in the search for the most parsimonious tree. Autapomorphies were not included in computation of tree statistics.

Multistate characters (3, 4, 18, 20, 22, 41, 45, 46, 51, 62, and 65) were first analyzed as unordered. Then those multistate characters having states that could be ordered sequentially according to their divergence from that putatively primitive condition were analyzed as ordered (3, 4, 18, 20, 41, and 51).

No specific optimization method, i.e., “accelerated transformation optimization” (ACCTRAN) or the “delayed transformation optimization” (DELTRAN), was used to eliminate equally parsimonious alternative hypotheses of character state transformations on a cladogram. Tree manipulations and diagnostics were done with the help of MacClade computer program, version 3 by Maddison and Maddison (1992), and Clados, version 1.2 by Nixon (1992).

Missing entries in the data matrix (represented by “?” in appendix 1) were employed in the present study for two distinct situations: (1) character state not checked due to lack of study material (only one instance represented by character 15 in Rhaphiodon vulpinus); and (2) character state inapplicable (coded in some taxa for characters 6, 7, 12, 20, 21, 29, 39, and 65). In order to understand the impact of characters represented by missing entries in the resulting phylogenetic hypothesis, various analysis were performed that included and excluded all of those characters or combinations of them.

Character polarity was determined by outgroup comparison and in one instance (character 39) by using ontogenetic information. Thirty-three characiform outgroup taxa listed above (indicated by stars following the species name) constituted the focus of these comparisons. These taxa were previously proposed as being related to cynodontines either as sister groups or at higher levels of inclusiveness based on the possession of some common features. In addition to the seven cynodontine species examined in the present study those outgroup taxa were checked for all examined characters (Cynodon meionactis was not included in the present phylogenetic analysis, see discussion under Comments on Cynodon meionactis, following the key to Cynodon species, below). In addition to these taxa, a number of other characiforms not directly related to cynodontines (listed above, without asterisk following species names) were examined to assess character distribution and variation within characiforms. The occurrence of character states hypothesized as derived for cynodontines in these more distantly related outgroups is discussed under the appropriate characters. Outgroups presented in the matrix (appendix 1) were restricted to Roestes, Gilbertolus, and Acestrorhynchus, hypothesized to be the closest relatives of cynodontines (Lucena and Menezes, 1998), plus a hypothetical ancestor that summarizes the observations made for remaining outgroup taxa. Justification for outgroup choice is presented under Historical Overview and Comments on Cynodontine Relationships with Other Characiforms.

The Character Description and Analysis section includes a description of each character, its variation within cynodontines, its occurrence in characiform outgroups, and its optimization in the resulting phylogenetic hypothesis. In a few instances characters were observed that show variation within cynodontines and could provide information on cynodontine intrarelationships; however, difficulty in the interpretation of their states complicated the proposition of hypothesis of homology (see Phylogenetic reconstruction for a list of those characters). Various analyses were performed—including, excluding, or testing different codings of such characters—to study their impact on the resulting phylogenetic hypothesis. Problematic characters mentioned above, characters pertinent to the question of cynodontine relationships to other characiforms, and autapomorphic characters are not included in the computation of tree statistics (44 out of the 69 characters were included in the computation of tree statistics), but are included in the Character Description and Analysis section. Some of the characters studied are concerned with the question of cynodontine relationships to other characiforms. Although the question of cynodontine relationships to other characiforms did not constitute the primary objective of the present study, discussions of these characters should be considered pertinent to the present analysis.

Terminology

Osteological terminology follows Weitzman (1962) with a few modifications proposed by various authors. Vomer is substituted for prevomer, epioccipital for epiotic, posterior ceratohyal for epihyal; anterior ceratohyal for ceratohyal, and mesethmoid for ethmoid.

Species Accounts

The taxonomic section of the study is based on analysis of meristic and morphometric characters. Counts and measurements were made on the left side of the specimens, except when the structure being measured or counted was recognizably abnormal or damaged, in which case corresponding data were taken from the right side. Measurements were taken with calipers and data recorded to tenths of a millimeter for distances under 150 mm and to a millimeter for larger distances. All measurements were taken point to point, i.e., not orthogonal to the main body axis. Vertebral counts and pterygiophore insertion relative to neural and hemal spines were examined from radiographs. Counts and measurements are presented in tables and/or in the text and follow Fink and Weitzman (1974) except as noted, with some additions given below: body depth at dorsal-fin origin—immediately anterior to dorsal-fin origin; body depth at pelvic-fin origin—immediately anterior to insertion of the unbranched pelvic-fin ray; dorsal-fin origin to caudal-fin origin—to center posterior termination of the hypural fan (where the principal caudal-fin rays attach to the hypural bones); dorsal-fin origin to adipose-fin origin; pectoral-fin origin to anal-fin origin; dorsal-fin base and anal-fin base—between anterior and posterior termination of fin-ray bases; interorbital width—between borders of frontal bones at anterior tip of supraorbitals; postorbital length—from posterior bony orbital margin to posteriormost termination of opercular bone (without including the fleshy opercular flap); upper jaw length—from tip of snout to distal tip of maxilla; dentary canine length—from base to tip of largest canine tooth in dentary; scales along lateral-line series—counted as longitudinal scale row along perforated lateral-line scales, including those on base of caudal-fin; scale rows below lateral-line—number of longitudinal rows of scales counted from anus to that scale just ventral to (and not including) the perforated lateral scale row. This form of counting scales below the lateral series is found to be more replicable than the traditional form, in which the count is made to the origin of the anal fin. Cynodontines have the scales below the lateral series obliquely arranged, becoming smaller and irregularly organized toward the region of the anal fin, rendering the count inaccurate when taken at the level of the anal-fin origin. Counts taken at the level of the pelvic-fin origin also showed a large degree of inaccuracy. Counting scale rows below the lateral series at the level of the anus, slightly in advance of the anal-fin origin, yielded more replicable counts although some inaccuracy remained. For gill-raker count the limit between the anterior limit of the ceratobranchial and posterior end of the hypobranchial was set at the anterior gill raker in the lower limb of the first gill arch which has its basal portion in contact with the dorsal end of the gill filaments of the same gill arch. Anterior to this point there is a gap between the basal portion of the gill raker and dorsal end of the gill filaments, so that they are not in contact but are separated by skin. When examined in cleared and stained specimens, this point largely corresponds to the limit between the cerato- and hypobranchial of the first gill arch. This system of counting gill rakers is found to be more replicable than counting all gill rakers on the lower limb of the first gill arch (i.e., on both ceratobranchial and hypobranchial) because the anterior rakers could not be easily reached without damage to the specimen (particularly large ones), and rakers on the hypobranchial tend to be smaller and fused to one another to different extents, making the counts susceptible to inaccuracy; vertebral counts: vertebrae incorporated into the Weberian apparatus were counted as four elements and the fused PU1 + U1 was considered a single bone. Measurements of anal-fin length were not included because the tips of the anterior fin rays were damaged in the majority of specimens.

In tables and text, subunits of the head are presented as proportions of head length. Head length and measurements of body parts are given as proportions of standard length. Numbers in parentheses following a particular vertebral count are the number of radiographed specimens with that count. In counts of fin rays, the unbranched fin rays are indicated by lower case roman numerals, and the branched fin rays as arabic numerals.

The Material Examined section of each species account is arranged in the following sequence: total number of specimens examined, and in parentheses, the number of specimens from which counts and measurements were taken, and their range of standard lengths (in mm). The lots are grouped according to country and, within each country, the state or department, followed by institutional abbreviation, catalog number, number of specimens in the lot and their range of standard lengths (the latter two in parentheses), and specific locality data. Institutional abbreviations follow Leviton et al. (1985) and Leviton and Gibbs (1988).

In the key to the species of Cynodon below, information on C. meionactis is provided by Géry et al (1999), and some of it was recorded differently from the morphometric data in the present study (e.g., orbital diameter) (see Comments on C. meionactis, below).

Orbital Region

1. Fifth infraorbital

Cynodontines have six canal-bearing bones forming the orbital ring with bony plates that cover part or all of the adductor musculature of the cheek (fig. 2), a condition considered to be primitive for characiforms (Roberts, 1969: 419; Vari, 1979: 301; Fink and Fink, 1981: 315). In Cynodon, Rhaphiodon, Hydrolycus scomberoides, and H. wallacei the fifth infraorbital typically reaches the posterior margin of the infraorbital series and the fourth and six infraorbitals are not in contact (fig. 2B). In Hydrolycus armatus and H. tatauaia the fifth infraorbital is greatly reduced, with the posteroventral margin of the sixth infraorbital in contact with the posterodorsal margin of the fourth infraorbital (fig. 2A).

Different patterns of reduction of infraorbital bones are observed in various characiform groups (Weitzman and Fink, 1983), some having only five infraorbitals (e.g., Charax, Lucena, 1987), and others having a reduced or absent fourth infraorbital (e.g., ctenolucids, Vari, 1995). The pattern described above for Hydrolycus armatus and H. tatauaia is, however, unique among characiforms, and considered synapomorphic for the clade formed by these two species.

2. Antorbital-lateral ethmoid contact

All cynodontines have the antorbital bone contacting the lateral margin of the ventral wing of the lateral ethmoid. The antorbital in characiform outgroups having this ossification (a separate antorbital is lacking in ctenolucids and erythrinids, see Vari, 1995: 9) is usually positioned anterior to the ventral wing of the lateral ethmoid with no, or only slight, contact between the dorsalmost portions of these two ossifications.

An extended contact between the antorbital and the lateral edge of the ventral wing of the lateral ethmoid occurs in Gasteropelecus and Agoniates. In the latter genus the antorbital is well developed (Géry, 1962) with the lateral edge of the ventral wing of the lateral ethmoid anteroposteriorly expanded and forming a distinct lateral surface instead of the narrow edge present in other characiforms. In addition to the antorbital-lateral ethmoid contact, gasteropelecids also share the presence of expanded coracoids with cynodontines. These two taxa do not seem, however, to be closely related as discussed by Weitzman (1954). Gasteropelecids and cynodontines differ significantly in most parts of their osteology, with the former hypothesized to be more closely related to characiforms such as Astyanax, Brycon, and Bryconamericus (Weitzman, 1954: 231). The relationships of Agoniates have been discussed in the previous section and the features shared by these taxa are hypothesized as having originated independently. The antorbital bone contacting the lateral margin of the ventral wing of the lateral ethmoid is, therefore, hypothesized as a synapomorphy for cynodontines.

3. Antorbital

The antorbital bone in all Hydrolycus species (fig. 3A) is a flat, platelike ossification without an extending process.

The antorbital in Rhaphiodon and Cynodon (fig. 3B) has a medial, vertically aligned process that extends along the posterior surface of the ventral wing of the lateral ethmoid. In Rhaphiodon this process is very narrow, extending only slightly medially from the posterior margin of the antorbital and contacting the posterior margin of the ventral wing of the lateral ethmoid. In Cynodon this process is further developed, extending to a greater degree medially, especially in the middle portion of the process.

Flat antorbitals are observed in most characiform outgroups, but in some, the antorbital is not a flat ossification, but presents some degree of elaboration. This was observed in Acestrorhynchus, Gasteropelecus, Chalceus, Iguanodectes, and Xenocharax. The overall shape of the antorbital in these taxa differs in various ways from that in cynodontines, and therefore does not seem to be directly comparable to that of cynodontines.

The medial process on the posterior margin of the antorbital in Cynodon and Rhaphiodon is hypothesized as derived, as opposed to the condition of a flat, platelike ossification found in Hydrolycus and widespread among characiforms outgroups, and is interpreted as synapomorphic for the clade formed by these two genera. The further enlargement of the antorbital process in Cynodon is hypothesized as a synapomorphy for that genus.

Neurocranium

The anterior portion of the neurocranium (mesethmoid-vomer region) of cynodontines is highly modified compared to other characiforms; many specializations are found in the articulation of the premaxilla, maxilla, and ectopterygoid. All ossifications in this area are held together by strong ligamentous tissue. A series of anterior neurocranial features provide considerable morphological information pertinent to the elucidation of the phylogenetic relationships of the Cynodontinae, as detailed below.

4. Mesethmoid

Hydrolycus species have the anterior portion of the mesethmoid, which forms an articular surface for the premaxillae, dorsoventrally expanded, and almost round in shape when examined in lateral view. In Hydrolycus scomberoides, H. armatus, and H. tatauaia, (figs. 4, 5) there is a further expansion of this anterior portion of the mesethmoid compared to H. wallacei (fig. 6). This feature was previously noted by Starks (1926: 165) for Hydrolycus scomberoides (H. pectoralis of that author).

In Cynodon and Rhaphiodon, the mesethmoid ends in a conical, spinelike process (figs. 7 and 8, respectively), the condition observed in most examined characiform outgroups and considered plesiomorphic (Weitzman, 1962: 19; Roberts, 1969: 405; Vari, 1979: 277). A vertically enlarged anterior mesethmoid was observed in Mylossoma and Roeboexodon guyanensis among examined outgroups. In the latter species the anterior enlargement of the mesethmoid is accomplished by a vertical expansion of the spine, and is continuous with an expansion of the ventral surface of the mesethmoid, a condition different from that in cynodontines. This vertical expansion of the anterior mesethmoid in Roeboexodon guyanensis results in a bony plate that provides a very broad articular surface for the premaxillae.

Elaboration of the anterior portion of the mesethmoid occurs in a few characiforms e.g., Citharinidae and Distichodontidae, but they represent different modifications from the one described above for Hydrolycus (see Vari, 1979: 278–279, for details). The condition observed in Hydrolycus is unique among examined outgroups and is most parsimoniously interpreted as a synapomorphy for that clade, with further anterior enlargement of the mesethmoid in Hydrolycus scomberoides, H. armatus, and H. tatauaia hypothesized as a synapomorphy for this clade.

5. Ventral process of the mesethmoid (sensu Starks, 1926: 163)

All cynodontines have a pair of processes projecting from the ventral surface of the mesethmoid (figs. 4–8). These processes arise from the ventral surface of the lateral wings of the mesethmoid to which the premaxillae are attached, and are interposed between the base of the mesethmoid spine anteriorly, and articulation with the vomer posteriorly. The processes bear cartilage pads on their ventral tips. Starks (1926: 163) reported the presence of the ventral processes on the mesethmoid in Rhaphiodon vulpinus and Hydrolycus scomberoides (H. pectoralis of that author). A sheet of thick connective tissue is attached to the tip of these processes and covers the entire ventral portion of the vomer-mesethmoid region. The layer of connective tissue is less dense posteriorly and attaches to the medial portion of the maxilla. In juveniles of Rhaphiodon (USNM 231549, 41.7–50.1 mm SL) the processes are represented by two large cartilaginous pads on the ventral surface of the ossified portion of the mesethmoid.

Determination of the element homologous to the ventral processes of the mesethmoid described for cynodontines in characiform outgroups is complicated. Among examined outgroups, an ossification similar to the ventral processes of the mesethmoid described for cynodontines occurs in erythrinids, Lebiasina bimaculata, and Piabucina sp. In these taxa the processes occupy the same position on the mesethmoid as the structures in cynodontines, originating ventrally from the lateral groove of the mesethmoid where the premaxillae attach. Such processes in erythrinids and lebiasinids are more laterally oriented than the processes in cynodontines, in which they lie somewhat parallel to each other. The posterior edges of the processes in erythrinids and examined lebiasinids contacts the anterior margin of the vomer (as in Rhaphiodon and some Hydrolycus species, detailed below). The sheet of connective tissue attached to the margins of these processes in cynodontines was absent in these outgroups. Rather, in the outgroups the cartilage at the tip of the ventral process of the mesethmoid contacts a cartilaginous surface on the anterior portion of the palatine, and ligamentous tissue suspends the maxilla and palatine arch to the ventral process of the mesethmoid. Weitzman (1964: 137) described the mesethmoid of Hoplias and Erythrinus as being similar to that of Brycon. The latter genus lacks the ventral processes of the mesethmoid noted for cynodontines. The two lateral processes in the mesethmoid of Brycon were described as the “lateral ethmoid wing” (= lateral wings of the mesethmoid) by Weitzman (1962: 19). Ventral parts of the lateral wings of the mesethmoid in Brycon contact the upper portions of the vomer via a synchondral joint (Weitzman, 1962: 19). As noted above, the ventral processess of the mesethmoid in erythrinids also join the anterior margin of the vomer by a synchondral joint. It is not clear whether Weitzman (1964: 137), in mentioning the similarity between the mesethmoid in Brycon to that in Hoplias and Erythrinus, was referring just to the similarity in overall shape or implied also that the lateral processes present in the mesethmoid of these taxa represented the same structure. A lateral wing of the mesethmoid was coded as present for Hoplias by Buckup (1991: 213, character 2). It is also not clear whether what was considered the lateral wing of the mesethmoid by Buckup is what it is called here the ventral processes of the mesethmoid or is rather the very narrow lateral extension of the mesethmoid dorsal to these processes where the ascending processes of the premaxilla are connected. In cynodontines a distinct structure can be recognized extending laterally from the mesethmoid onto which the premaxilla and the anterodorsal process of the maxilla attach; that structure seems to correspond to the lateral wing of the mesethmoid as described by Weitzman (1962: 19). A similar element is absent in erythrinids.

Processes on the ventral portion of the mesethmoid bearing some resemblance to those in cynodontines were found in Prochilodus rubrotaeniatus, Caenotropus maculosus, Laemolyta taeniata, and Hemiodus sp. among examined outgroups. In the latter two species the anterodorsal portion of the vomer is separated from the ventral portion of the mesethmoid processes by a large cartilage block. In Prochilodus rubrotaeniatus and Caenotropus maculosus the posterior portions of the processes contact the anterior portion of the vomer and laterally they contact the anterior cartilaginous surface of the palatine. Weitzman (1954: 217) described a ventral process on the mesethmoid contacting the vomer along a broad suture in Carnegiella vesca. This process observed here in Carnegiella strigata, bears little resemblance to the processes noted above for cynodontines and some characiform outgroups.

Starks (1926: 163), in describing the ventral processes of the mesethmoid in Rhaphiodon vulpinus, stated that “These are the same processes described for Hoplias and doubtless the same as those of the cyprinoids that bear the pre-ethmoids.” In the description of the ethmoid region of Hydrolycus pectoralis (H. scomberoides) Starks (1926: 165) stated that “The pre-ethmoid processes of the mesethmoid are the same.” [as in Rhaphiodon vulpinus]. For Distichodus fasciolatus Starks (1926: 169) noted that “A nodule of cartilage filling a concavity at the union between the mesethmoid and the vomer doubtless represents the pre-ethmoid.” Fink and Fink (1981: 312), in their discussion of ostariophysan interrelationships, mentioned that (their character 4): “In cypriniforms a cartilage body or endochondral ossification, usually termed the ‘pre-ethmoid’, is tightly articulated between the vomer and mesethmoid. In Chanos, many characiforms [my emphasis], and some other teleosts a probably homologous cartilaginous or ossified body is present between the palatine, maxilla, and ethmoid.”

The available evidence precludes an adequate assessment of homologies of the ventral processes of the mesethmoid for cynodontines relative to the condition found in other characiforms. However, all the groups listed above in which relatively similar processes were observed are hypothesized as being more closely related to groups that lack well-developed processes. Prochilodus rubrotaeniatus, Caenotropus maculosus, and Laemolyta taeniata belong to three different monophyletic families that together with the Curimatidae (in which such ventral processes on the mesethmoid are lacking) are hypothesized to form a monophyletic group (Vari, 1983). Erythrinids and lebiasinids share many derived features with Hepsetus and ctenolucids, taxa that also lack these processes. All the groups that are potentially related to the Cynodontinae at different levels of inclusiveness (Roestes, Gilbertolus, and Acestrorhynchus—see discussion on section about cynodontine interrelationships) lack processes on the ventral portion of the mesethmoid that resemble to those described for the Cynodontinae. As a consequence, the ventral processes on the mesethmoid as described for the Cynodontinae are hypothesized as synapomorphic for the members of that subfamily.

6. Orientation of ventral processes of the mesethmoid

In all cynodontines except Hydrolycus wallacei, the ventral processes of the mesethmoid are ventrally directed at an approximately 90° angle relative to the mesethmoid spine (figs. 4, 5, 7, 8). In Hydrolycus wallacei (fig. 6), however, these processes are reoriented into a forward direction at a considerably smaller angle relative to the mesethmoid spine.

This character was coded as a missing entry (meaning inapplicable) in the outgroups since they do not possess ventral processes on the mesethmoid as described for cynodontines. According to the most parsimonious hypotheses of cynodontine intrarelationships generated in the present study, this character is interpreted as an autapomorphy for Hydrolycus wallacei.

7. Vomer–mesethmoid contact

In Hydrolycus tatauaia, H. armatus, H. scomberoides, and Rhaphiodon vulpinus, the anterior surface of the vomer contacts the ventral process of the mesethmoid along its posterior surface (figs. 4, 5, 8). In Cynodon (fig. 7) the vomer does not directly contact the ventral processes of the mesethmoid; its contact occurs dorsal to the posterior surface of the ventral processes of the mesethmoid, which are free of contact with other ossification. The condition in Hydrolycus wallacei is difficult to interpret due to the anterior shift of the ventral process of the mesethmoid (charater 6), and although the vomer does not contact that process along its entire posterior surface as in the remaining Hydrolycus species and Rhaphiodon, there is contact along the dorsalmost portion of the processes, a condition slightly different from that in Cynodon. For this reason the condition in Hydrolycus wallacei was coded as a missing entry.

This character was also coded as a missing entry in outgroups since they lack the ventral processes on the mesethmoid. All examined outgroups that have an element similar to the ventral process of the mesethmoid described for cynodontines, have that process contacting the anterior surface of the vomer (e.g. in erythrinids), a condition similar to that in Hydrolycus tatauaia, H. armatus, H. scomberoides, and Rhaphiodon vulpinus. However, as noted under character 5 above, homology propositions for the ventral processes of the mesethmoid in cynodontines relative to the condition found in characiform outgroups are not clear, and all taxa that possess similar processes are more closely related to taxa that lack them.

Examination of different ontogenetic stages did not provide information that could be used to determine polarity of this character. In juveniles of Rhaphiodon the anterior portion of the vomer is already in contact with the posterior portion of the ventral processes of the mesethmoid, which are cartilaginous at this stage (USNM 231549, 41.7–50.1 mm SL). It was not possible to examine this feature in juveniles of Cynodon due to the poor condition of the specimen available (AMNH 32485, 35 mm SL).

The distribution of this character in cynodontines renders any kind of optimization for the character states arbitrary. As a consequence it is impossible to unambiguously propose this character as an additional synapomorphy for any clade under the phylogenetic reconstruction summarized in the cladogram (fig. 20). Tests including and excluding this character from the analysis did not change the resulting hypothesis.

Some modifications of the anterior portion of the neurocranium in cynodontines are associated with the area of articulation with the maxilla. Such modifications occur on the lateral portion of the vomer, lateral wing of the mesethmoid, ventral diverging lamellae of the mesethmoid, and in part, the ventral processess of the mesethmoid. In cynodontines those elements delimit a groove or fossa that provides an area of attachment for the ligamentous tissue of the medially directed anterodorsal process of the maxilla. The series of modifications that contribute to form this fossa and the variation exhibited within cynodontines are detailed under characters 8–10 below.

8. Ventral diverging lamellae of the mesethmoid

Cynodontines have well-developed ventral diverging lamellae of the mesethmoid (sensu Weitzman, 1962: 19). Posteriorly, the lamella of each side contacts the upper central portion of each lateral ethmoid. Anteriorly the lamella of each side is continuous with a portion of the lateral wing of the mesethmoid medial to the region where the maxilla attaches, and extends posteriorly on the lateral surface of the vomer. This results in a well-developed triangular bony plate, with one corner of the triangle pointing toward the posterior margin of the vomer (figs. 5–8) (henceforth the entire element will be referred to as the diverging lamella of the mesethmoid). Together the triangular bony plates from each side clasp the lateral surface of the vomer. The ventral margin of this triangle forms the dorsal limit of a fossa where the maxilla articulates.

Hydrolycus scomberoides has a slightly different condition from the one described above. In this species there is no continuity of the lamella along the ventral surface of the mesethmoid with the portion of the lamella extending from the lateral wing of the mesethmoid (fig. 4).

The diverging lamellae of the mesethmoid are reduced or absent in many characiform outgroups, e.g., erythrinids, lebiasinids, ctenoluciids, and gasteropelecids. Weitzman (1962: 137) stated that the ventral diverging lamellae of the mesethmoid are well developed in the Erythrinidae; however, I agree with Buckup (1991: 214) who regards Weitzman's mention of such lamellae as a misinterpretation of a similar ossification that extends posteriorly from the dorsal surface of the lateral mesethmoid wings. A few Acestrorhynchus species (A. falcirostris and A. nasutus) have a lateral lamella that extends ventrally from the mesethmoids with its anterior portion contacting the vomer and extending posteriorly toward the lateral ethmoid. In the remaining Acestrorhynchus species, such a lamella is very reduced and does not reach the vomer ventrally. As in the Erythrinidae it seems that the lamella as described for Acestrorhynchus species does not correspond to the ventral diverging lamella of the mesethmoid described by Weitzman (1962: 19). Characiforms with well-developed ventral diverging lamellae of the mesethmoid include Roestes, Gilbertolus, Charax, Roeboexodon, the Cynopotaminae (Menezes, 1976), Acanthocharax, and Agoniates.

A discontinuity between the anterior and posterior portion of the diverging lamella of the mesethmoid is restricted to Hydrolycus scomberoides within cynodontines, a condition most parsimoniously interpreted as autapomorphic for this species.

9. Triangular portion of the ventral diverging lamellae of the mesethmoid

The extent to which the triangular portion of the ventral diverging lamella of the mesethmoid (described under character 8) extends along the lateral surface of the vomer varies among cynodontines. In all Hydrolycus species and Rhaphiodon the triangular portion of the ventral diverging lamella of the mesethmoid extends laterally over the anterodorsal corner of the vomer, leaving most of the lateral surface of the vomer uncovered (figs. 4–6, 8). In Cynodon this lamella is more elongate, extending beyond the midline, and in many specimens reaching the posteroventral edge of the vomer (fig. 7). As a consequence, in Cynodon most of the lateral surface of the vomer is covered by the triangular portion of the ventral diverging lamella of the mesethmoid.

At first a direct comparison of the condition of this character in outgroups with that in cynodontines seemed complicated, in view of the high degree of modification of the entire vomer-mesethmoid region in cynodontines. However, closer examination of this feature in outgroups revealed that in some, the portion of the lamella at the ventral surface of the mesethmoid that extends ventrally to contact the anterior portion of the vomer is continuous with the portion extending medially and posteriorly from the lateral wing of the mesethmoid (Charax, Acanthocharax, Galeocharax, Brycon, Roeboexodon). This portion of the ventral diverging lamella of the mesethmoid in examined outgroups does not extend as far posteriorly on the surface of the vomer, in many cases being very reduced or lacking (e.g., Roestes, Gilbertolus, and Agoniates). Therefore, the triangular portion of the ventral diverging lamella of the mesethmoid covering most of the lateral surface of the vomer is hypothesized as synapomorphic for Cynodon.

10. Articular surface on lateral wing of the mesethmoid

The region of the lateral wing of the mesethmoid where part of the ligamentous tissue of the maxilla attaches differs among cynodontines. In Hydrolycus tatauaia, H. scomberoides, and H. armatus this portion of the lateral wing of the mesethmoid has a distinct posteriorly oriented articular surface (figs. 4, 5). Ligamentous tissue from the maxilla converges anteriorly to attach to this articular surface, which forms the anterior boundary for a depression on the lateral surface of the vomer where the maxilla abuts at the level of the synchondral joint between the vomer and ventral processes of the mesethmoid (see character 14 below). In Cynodon, Rhaphiodon, and Hydrolycus wallacei, ligaments from the maxilla attach to the ventral portion of the lateral wing of the mesethmoid without a distinct posteriorly directed articular surface forming an anterior boundary for the area of articulation of the maxilla. As a consequence the depression on the lateral surface of the vomer where the maxilla abuts extends slightly more anteriorly and reaches the posterolateral surface of the mesethmoid. This is more evident in Hydrolycus wallacei and Rhaphiodon (figs. 6 and 8, respectively) than in Cynodon (fig. 7) in which the depression on the vomer is reduced compared to the remaining species.

Ligamentous tissue suspending the upper anterior tip of the maxilla to the lateral wing of the mesethmoid is the typical condition among characiforms (Weitzman, 1962: 19; Roberts, 1969: 405), and also occurs in cynodontines. Comparisons of the portion of the mesethmoid where the maxilla attaches in characiform outgroups and cynodontines is complicated due to the highly modified mesethmoid region of cynodontines. The lateral wings of the mesethmoid in many characiforms outgroups have a laterally oriented cartilaginous cap that is continuous with the cartilaginous surface from the lateral portion of the vomer (e.g., Roestes, Charax, Galeocharax, Heterocharacini), and from which the maxilla is suspended by ligaments. In Brycon, Acestrocephalus, Acanthocharax, and Acestrorhynchus the ligamentous tissue from the anterior tip of the maxilla attaches to a small portion of the lateral wing of the mesethmoid that has a ventral to slightly lateral articular surface.

A posterior orientation of the articular surface of the lateral wing of the mesethmoid onto which part of the ligamentous tissue of the maxilla attaches is unique to Hydrolycus tatauaia, H. scomberoides, and H. armatus among examined characiforms; this seems to be the result of the reorientation of the lateral wing of the mesethmoid to a somewhat more vertical position, shifting the surface of the articulation of the ligaments of the maxilla to a more posterior orientation. Such a reorientation of the lateral wing of the mesethmoid is also expressed in the orientation of its articular surface with the premaxilla that is more closely associated with the lateral portion of the ventral processes of the mesethmoid in Hydrolycus tatauaia, H. scomberoides, and H. armatus compared to the condition in Cynodon and Hydrolycus wallacei (in Rhaphiodon the articular surface of the lateral wing of the mesethmoid with which the premaxilla articulates is less pronounced laterally). The condition observed in Hydrolycus tatauaia, H. scomberoides, and H. armatus is hypothesized as a synapomorphy for that assemblage.

The vomer in cynodontines is a massive bone that fills the space between the mesethmoid and lateral ethmoids. It is highly concave ventrally and dorsolaterally, being an inverted Y-shaped bone posteriorly in cross section (see character 13 below) with the angle between the two arms of the Y very acute, and the vertical process that forms a sagittal bony plate elongate, and contacting a similar bony plate of the mesethmoid anteriorly.

An inverted Y-shaped vomer was described for the Erythrinidae and Lebiasinidae by Weitzman (1964: 138). The angle between the two arms of the Y in these families is not, however, as acute as in cynodontines. In erythrinids and lebiasinids the dorsolateral surface of the vomer forms an almost a 90° angle, whereas in cynodontines this angle is much greater. A vomer like that in cynodontines was observed only in Distichodus fasciolatus, among examined characiforms, with the overall shape of this ossification being more similar to that of Cynodon. Modifications of the vomer that provide information about cynodontine intrarelationships are described below.

11. Ventral crest of vomer

In Hydrolycus species and Rhaphiodon the ectopterygoid is in close contact with the ventral surface of the vomer and not with the mesethmoid as in Cynodon (character 31 below). A close contact between the vomer and the ectopterygoid also occurs in Acestrorhynchus and Hepsetus but represents a different set of modifications than those found in the Cynodontinae.

The ventral surface of the vomer in Hydrolycus species and Rhaphiodon bears two longitudinal crests (figs. 4, 6, and 8, lateral view) situated side by side and separated by a longitudinal groove. The ectopterygoid of each side abuts the ventrolateral portion of each crest, which serves as an area of attachment for the ligamentous tissue between these two elements. This crest is developed to varying degrees among cynodontines when present. It is relatively small in Rhaphiodon and Hydrolycus wallacei and very developed in Hydrolycus scomberoides. A crest on the ventral surface of the vomer is lacking in Cynodon.

In the majority of examined characiforms (with the exception of Acestrorhynchus and Hepsetus, see character 31) the ventral surface of the vomer is flat, with no conspicuous elaborations for the attachment of ligaments. In Agoniates a median ridge that delimits two fossa for the attachment of ligaments arises from the mesopterygoid. Gilbertolus atratoensis possesses two small projections of bone on the ventral surface of the vomer onto which ligamentous tissues from the mesopterygoid and ectopterygoid attach bearing some resemblance to those described for Hydrolycus and Rhaphiodon. The ectopterygoid does not directly contact these processes in Gilbertolus atratoensis as it does in Hydrolycus and Rhaphiodon. In the specimen of Roestes ogilviei examined, the processes are hardly noticeable, and they are absent in the examined specimen of R. molossus.

Interpretation of the ventral processes on the vomer of Hydrolycus and Rhaphiodon is ambiguous under all optimizations of this character. In addition, the slight resemblance of the condition in those cynodontine taxa to that in Gilbertolus further complicates any conclusion about the evolution of this feature. Therefore, this character was not hypothesized as synapomorphic for the species that possess it. Analyses that included, excluded, or used different character codings of this character did not affect the resulting phylogenetic hypothesis.

12. Lateral arms of vomer

In all cynodontines the two lateral bony plates forming the two arms of the inverted Y of the vomer are well developed along the posterior portion of the ossification. Some variation in the anterior portion of the vomer is, however, observed among cynodontines. In Cynodon these two lateral bony plates are continuous all the way to the anterior portion of the vomer, nearly as far as the articulation with the mesethmoid. As a consequence, in Cynodon the vomer in cross section is an inverted Y along its entire length. In all Hydrolycus species and Rhaphiodon the lateral arms of the vomer gradually diminish in relative size anteriorly, extending only onto the posterior half of the bone. As a consequence, from a ventral view, the concavity of the vomer is only evident posteriorly. Anteriorly the vomer in Hydrolycus species and Rhaphiodon is not an inverted Y in cross section but rather flat ventrally. As already mentioned, the vomer of Distichodus fasciolatus, is similar to that of Cynodon, with continuous bony plates along its entire length.

In characiform outgroups that have a shape of the vomer similar to that in cynodontines (e.g., erythrinids, lebiasinines), and those in which the shaft of the vomer has a relatively constant width along its length (e.g., Hoplocharax, Heterocharax, gasteropelecids) the lateral portion of the vomer usually extends beyond the longitudinal axis of the shaft of the vomer and is continuous along the length of the vomer. In other characiforms the vomer is a relatively flat ossification that extends laterally to varying degrees beyond the margin of the shaft. In most of these groups (e.g., Roestes, Gilbertolus, Acestrocephalus, Charax, and Brycon), there is an enlarged articular surface at the anterolateral portion of the vomer that widens this portion of the bone. This articular surface is lacking in cynodontines. A direct comparison of the condition in these taxa to that in cynodontines is, as a consequence, further complicated. The condition described for Cynodon, is most parsimoniously optimized as derived in the cladogram that resulted from the analysis but it is not at this time proposed as an additional synapomorphy for that genus since the condition in outgroups was coded as a missing entry.

13. Vomer-palatine contact

In all cynodontines the anterior portion of the palatine has a cartilaginous articular surface that abuts a matching articular facet in the posterior edge of the main body of the vomer.

Contact between the vomer and the palatine is relatively common among characiforms. What differs, however, is the type of contact. The anterior cartilaginous surface of the palatine in characiforms typically contacts the anterolateral portion of the vomer by way of a relatively loose type of contact. In ctenoluciids the articular surface of the vomer is situated laterally between the anterior and posterior edges. Acestrorhynchus species seem to have a unique condition among characiforms in which the anterior portions of the palatine abuts against two ventral processes in the ventral surface of the vomer. In erythrinids the palatine contacts a ventral process in the mesethmoid through ligamentous attachments. Prochilodus has two large articular surfaces located somewhat posteriorly on the vomer in contact with the palatine. However, the shape of the vomer in the latter genus is highly modified relative to that in most characiforms and this makes a direct comparison with the condition observed in cynodontines difficult. Roeboexodon also has two large articular surfaces in the vomer that contact the palatine. These articular surfaces although somewhat posterior in the vomer are oriented ventrally, and thus, are not directly comparable to the condition in cynodontines.

An articulation between the vomer and the palatine of the type observed in cynodontines also occurs in the Serrasalminae among examined characiforms and has been hypothesized as a synapomorphy for the latter group (Machado-Allison, 1983: 163). The phylogenetic relationships of the Serrasalminae remain unresolved. Machado-Allison (1983) discussed the problem and mentioned characters in common between serrasalmines and Brycon. Lucena's (1993) study on relationships within the Characidae presented evidence supporting a close relationship between serrasalmines, Chalceus, Brycon, and African characids, taxa lacking such vomer-palatine contact. In Hepsetus there is an elongate cartilaginous region at the posterior edge of the vomer that the cartilaginous surface of the anterior portion of the palatine contacts, a condition similar to that in cynodontines. Hepsetus is hypothesized as being closely related to the Ctenoluciidae and Erythrinidae, families with a different type of vomer-palatine contact. The vomer-palatine articulation found in these characiforms outgroups is, therefore, most parsimoniously cinsidered as originating independently from that in cynodontines, and therefore, the condition in the latter taxa is hypothesized as synapomorphic.

14. Vomer-maxilla articulation

The maxilla of cynodontines articulates with the lateral surface of the vomer at the anteroventral portion of the latter ossification. In this area of contact the vomer has a shallow depression. Anterodorsally this depression is roofed by the anteroventral portions of the ventral diverging lamella of the mesethmoid (see character 9 above), and the posteroventral portions of the lateral wing of the mesethmoid. The latter serves as an area for ligamentous attachment of the maxilla to the neurocranium (see character 10 above).

In Hydrolycus species, the depression on the vomer is more accentuated than in Cynodon and Rhaphiodon, with the latter exhibiting a condition that is intermediate between that of Cynodon and Hydrolycus. This feature demonstrates ontogenetic variation. In a small specimen of Rhaphiodon (BMNH 1935.6.4: 33–39, 96.8 mm SL) the depression is absent, whereas in a larger individual (MZUSP 32812, 191.0 mm SL) it is obvious. In very large specimens of Hydrolycus armatus the depression in the vomer is developed and deep, forming a fossa (examined in a number of dry skeletons, e.g., AMNH 91344SD, 400 mm SL).

The differing degrees of development of the lateral depression in the vomer may be of phylogenetic interest among cynodontines. The ontogenetic variation observed within species, and the continuous variation among species, complicates the definition of discrete character states, making their definition impossible at this time. However, the presence of a varying developed depression on the lateral surface of the vomer associated with the articulation of the maxilla was not encountered outside the Cynodontinae among characiforms and, therefore, is hypothesized as an additional synapomorphy for the subfamily.

15. Ridge on lateral surface of vomer

In large specimens of Hydrolycus the lateral surface of the vomer develops a ridge that interrupts the continuity of the concavity on the surface of the bone. Such a ridge is already developed in specimens of Hydrolycus scomberoides, H. wallacei (fig. 6), and H. tatauaia of approximately 150 mm SL. In H. armatus it is absent in a specimen 136 mm SL (LACM 43295-36), and in a specimen 151.0 mm SL (MZUSP 32607) is only very slightly developed, represented by a thickening of bone in the area where the ridge is present in larger specimens. It is well developed in an H. armatus specimen of 200 mm SL (MZUSP 32607). It was not possible to determine whether such a ridge is present in Rhaphiodon. In the largest Rhaphiodon specimen prepared for osteological examination (MZUSP 32812, 191 mm SL), a thickening of bone is observed slightly posterior to where the tip of the ventral diverging lamellae of the mesethmoid contacts the vomer. However, it is not continuous along the lateral surface of the bone as in other cynodontines with the ridge. Larger specimens of Rhaphiodon vulpinus have to be examined to further investigate the presence of such a ridge in this species. Therefore, this character was, coded as a missing entry for the latter species.

The lateral ridge of the vomer in Hydrolycus species forms a link between the tip of the triangular portion of the ventral diverging lamella of the mesethmoid anteriorly, and the process that extends from the lateral ethmoid posteriorly. This entire portion becomes massively developed in very large specimens of Hydrolycus (observed in dry skeletons of H. armatus), forming a continuous bridge extending from the mesethmoid to the lateral ethmoid. As a consequence, the median vertical plate of the vomer that is continuous with the remaining lateral surface of the vomer in small specimens seems almost a separate ossification occupying a more internal medial position, in the large specimens.

No such ridge on the lateral surface of the vomer was observed in any of the specimens of Cynodon. A bridge uniting the lateral ethmoid and mesethmoid is also present in Cynodon; however, it is formed by extensions of these two elements that are in direct contact with each other (in Hydrolycus these two elements are not in direct contact).

A distinct ridge on the dorsolateral surface of the vomer occurs in Agoniates and Distichodus fasciolatus among examined characiforms. This ridge is also in contact anteriorly with a process extending from the mesethmoid and posteriorly with a process from the lateral ethmoid. Presence of such a ridge on the lateral surface of the vomer is hypothesized as derived.

The relationships of Agoniates have been briefly discussed in the previous section. Distichodus is related to members of the Distichodontidae and Citharinidae (Vari, 1979), which lack the ridge on the vomer. The presence of the ridge on the vomer is most parsimoniously hypothesized as an independent acquisition in these outgroups.

The ridge on the dorsolateral surface of the vomer cannot be unambiguously hypothesized as synapomorphic for Hydrolycus at this time, pending a better definition of the condition in Rhaphiodon.

16. Rhinosphenoid

All Hydrolycus species and Rhaphiodon have a rhinosphenoid (figs. 4–6, and 8). Such a distinct ossification is absent from Cynodon.

The rhinosphenoid was first named and described by Starks (1926: 164) for Rhaphiodon vulpinus. In that species and all Hydrolycus species except H. scomberoides, the rhinosphenoid is a flat, and widely curved rod. Anteriorly it attaches to the midsagittal portion of the lateral ethmoids. Posteriorly it connects the medial portion of the orbitosphenoid (fig. 6). Ventrally, it contacts the parasphenoid through cartilage. It extends dorsally to contact the roof of the braincase just anterior to the fontanel. The dorsal extension of the rhinosphenoid forms a septum between the olfactory nerves as they issue from the anterior portion of the orbitosphenoid. In large specimens of these taxa the dorsolateral portion of each side of the orbitosphenoid develops a process that extends anteriorly and medially to meet the dorsal portion of the rhinosphenoid. As a consequence, two anterior openings for the olfactory nerve are delimited in the orbitosphenoid. The dorsal extension of the rhinosphenoid develops later ontogenetically than the remaining ossification. Specimens of Hydrolycus armatus smaller than 151.0 mm SL (fig. 5) and Rhaphiodon vulpinus of 96.8 mm SL (BMNH 1935.6.4: 33–39) have only the ventral portion of the ossification developed. Larger specimens (approximately 200.0 mm SL) have the dorsal extension of the rhinosphenoid well developed (compare figs. 5 and 6). A specimen of Hydrolycus tatauaia (MZUSP 32630, 151.2 mm SL) has the dorsal extension of the rhinosphenoid well developed.

In H. scomberoides the rhinosphenoid is reduced compared to that of the remaining cynodontines with this ossification (fig. 4). The dorsal extension is lacking in all specimens prepared for osteological examination, the largest being 241.0 mm SL (AMNH 40087SD, dry skeleton). The ventral portion of the rhinosphenoid in H. scomberoides is also reduced, and consists of a small curved rod filling the space between the orbitosphenoid and lateral ethmoids immediately dorsal to the parasphenoid.

The rhinosphenoid has been reported only among characiforms. Within the order, in addition to cynodontines, it also occurs in many lineages of the Characinae (sensu Weitzman, 1962), such as Acestrorhynchus, Roestes, Gilbertolus, Agoniates, Brycon, and others.

In Acestrorhynchus the rhinosphenoid is in contact with the parasphenoid, a feature that has been considered unique for the genus (Menezes and Géry, 1983). In small specimens this contact is made through cartilage between the two ossifications, whereas in larger specimens the bones are in direct contact. In Hydrolycus species and Rhaphiodon vulpinus the rhinosphenoid also contacts the parasphenoid through cartilage, with direct contact between the bones in larger specimens.

The absence of a rhinosphenoid is considered derived for Cynodon, with the absence of this ossification in other characiform lineages hypothesized as independent events.

17. Lateral ethmoid–orbitosphenoid contact

Cynodon, Rhaphiodon vulpinus, and Hydrolycus scomberoides have a type of lateral ethmoid–orbitosphenoid contact in which the dorsomedial portion of the lateral ethmoid bears a process extending posteriorly and contacting the anterodorsal portion of the lateral edge of the orbitosphenoid (fig. 7). This process originates dorsal to the olfactory foramen through the lateral ethmoid.

In small specimens of these taxa the anteromedian opening of the orbitosphenoid is open to the orbital cavity and it is covered dorsolaterally by the contact between the lateral ethmoid and orbitosphenoid. In larger specimens of Cynodon gibbus (MZUSP 32593, 179.0 mm SL, cleared and stained; AMNH 93079SD, 240.0 mm SL, dry skeleton), the contact between the lateral ethmoid and orbitosphenoid is enlarged by the dorsal growth of these two elements. As a consequence of this expansion and the growth of the ventral lamella of the frontal, the dorsolateral portion of the anteromedian opening of the orbitosphenoid is completely covered. The ventral portion of the lateral ethmoid process extends medially to almost meet the medial outgrowth of the same process of the contralateral ossification (I was unable to determine whether or not the process from each side meets ventromedially). The anteromedian opening for the olfactory nerve is restricted to the ventral portion of the orbitosphenoid. The olfactory foramen of the lateral ethmoid is positioned entirely ventral to the ventromedial portion of the lateral ethmoid process. It appears that the orbital cavity through which the olfactory nerve extends is restricted to a ventral position between the orbitosphenoid and lateral ethmoid.

Contact between the lateral ethmoid and orbitosphenoid is considered a derived condition in characiforms (Vari, 1979: 279) and it seems to have originated independently in the various groups where it occurs, i.e., the African characiforms Distichodontidae, Citharinidae, some Alestes species, Bryconaethiops, Hydrocynus, and the Neotropical groups Anostomidae, Curimatidae, Prochilodontidae, Lebiasinidae, and Parodontidae (Vari, 1979: 279–283). A type of lateral ethmoid-orbitosphenoid contact similar to that of cynodontines occurs in Piabucina, Lebiasina bimaculata, and Pyrrhulina sp. but is less developed in the latter two species.

The contact between the orbitosphenoid and lateral ethmoid in most groups mentioned above is of a different type than that in cynodontines. In distichodontids the lateral ethmoid process that contacts the orbitosphenoid originates from the ventral portion of that element (Vari, 1979: 280). The dorsal contact between these two elements in citharinids that is similar to that in cynodontines, but a second articulation between them is formed mainly by an anterior extension of the orbitosphenoid (Vari, 1979). In the Alestinae and Hydrocynus the process between the lateral ethmoid and orbitosphenoid is formed entirely by the latter element and is a bony tube surrounding the olfactory nerve (Roberts, 1969: 441; Brewster, 1986: 191). In Neotropical characiforms, contact between the lateral ethmoid and orbitosphenoid is made by a dorsal and ventral process. In all groups mentioned above, the processes between the lateral ethmoid and orbitosphenoid are associated with varying degrees of coverage of the olfactory bulb and tract.

The contact between the lateral ethmoid and orbitosphenoid is most parsimoniously hypothesized as derived for the clade formed by Cynodon and Rhaphiodon vulpinus with an independent origin in Hydrolycus scomberoides.

18. Orbitosphenoid-parasphenoid distance

All cynodontines have the orbitosphenoid in direct contact with the parasphenoid. However, the distance between the main portion of the orbitosphenoid and the parasphenoid differs among them. Hydrolycus wallacei, H. tatauaia, and H. armatus have the main portion of the orbitosphenoid well separated from the parasphenoid, with an elongate bony plate extending from the orbitosphenoid ventrally to contact the parasphenoid (figs. 5, 6). In Cynodon and Rhaphiodon vulpinus the distance between the parasphenoid and orbitosphenoid is greatly reduced and these two elements are in contact just ventral to the main body of the orbitosphenoid (figs. 7, 8). Hydrolycus scomberoides has an intermediate condition in which the bony plate contacting these two elements is not as dorsoventrally elongate as in the remaining Hydrolycus species (fig. 4), but is relatively more developed than in Cynodon and Rhaphiodon.

Many characiforms have the parasphenoid and orbitosphenoid remote from each other (e.g., Brycon, Roestes, Gilbertolus, Charax, the Heterocharacini, Acestrocephalus, Acestrorhynchus, and Serrasalmus). Such a separation was listed as one of the distinguishing features for the Characinae (Weitzman, 1962: 48), and also occurs in the Hemiodontinae (Roberts, 1974: 416). An orbitosphenoid close to the parasphenoid occurs in the Ctenoluciidae, Erythrinidae, Lebiasinidae, Hepsetus (Roberts, 1969: 406), Galeocharax, and the prochilodontids Semaprochilodus and Prochilodus (Roberts, 1973: 217). Galeocharax is more closely related to characids such as Acestrocephalus (Menezes, 1976) and Charax (Lucena, 1993); Semaprochilodus and Prochilodus are part of a monophyletic assemblage formed by the Curimatidae, Prochilodontidae, Anostomidae and Chilodontidae (Vari, 1983), all with well-separated parasphenoids and orbitosphenoids. The phylogenetic relationships of the clade formed by the Ctenoluciidae, Erythrinidae, Lebiasinidae, and Hepsetus (Vari, 1995) have not been the subject of investigation. However, in view of the current evidence for cynodontine relationships with Gilbertolus, Roestes, and Acestrorhynchus (see Historical Overview and Comments on Cynodontine Relationships with Other Characiforms), all of which have the orbitosphenoid well separated from the parasphenoid, the condition in Cynodon and Rhaphiodon is hypothesized as derived.

The orbitosphenoid close to the parasphenoid is hypothesized as a synapomorphy for the clade formed by Cynodon and Rhaphiodon, and the intermediate condition present in Hydrolycus scomberoides is considered autapomorphic for that species.

19. Dilatator fossa

The dilatator groove in cynodontines has some modifications relative to the condition in other characiforms. These modifications are most probably correlated with the origin of the dilatator operculi muscle in the orbit. The cranial musculature of cynodontines was studied by Howes (1976) who noted that in all cynodontine genera the dilatator operculi muscle has two origins. One is dorsally from the frontal-sphenotic groove, which extends onto the cranial roof, and the other is ventrally from the deep cavity that lies between the frontal and the orbitosphenoid.

The typical form of the dilatator groove in characiforms is one in which the frontal is strongly indented where it overlies the sphenotic, providing a sharp dorsal rim for the dilatator groove. The dilatator operculi originates from the fossa formed by the frontal and sphenotic, and in many cases by the lateral border of the pterotic (Roberts, 1969: 410 for Brycon, with discussion of variation of this pattern; Howes, 1976). In characiforms with this pattern, the dilatator muscle does not extend anteriorly onto the dorsal surface of the cranial roof past the posterodorsal margin of the orbit. (Note: different patterns of the dilatator groove occur in the Ctenoluciidae, Erythrinidae, and Acestrorhynchus and they represent a set of modifications different from those observed for cynodontines and will not be considered here. See Vari, 1995: 13–15 for details on the patterns the dilatator groove in these taxa.) Only in cynodontines is the dilatator fossa greatly expanded anteriorly, covering most of the dorsal surface of the frontal. An expanded dilatator groove extending onto the dorsal surface of the frontal at the middle of the orbit was reported for Cynopotamus (Menezes, 1976: 6). Cynodontines, as described above, have an even more developed dilatator fossa, a condition unique among characiforms and considered synapomorphic.

Different modifications of the neurocranium associated with the expansion of the dilatator fossa are observed within cynodontines. One modification occurs at the posterodorsal edge of the orbit, where the frontal and the dorsal portions of the sphenotic spine overlap to provide the dorsal rim for the dilatator groove. In the usual condition among characiforms, the dorsal portion of the sphenotic spine contacts the ventrolateral margin of the frontal, forming a continuous rim for the dilatator fossa that is developed laterally to varying degrees. This rim is very reduced in cynodontines as a function of two main modifications of this portion of the neurocranium. The first one is a function of the development of the frontal shelf at the posterodorsal edge of the orbit (character 20 below). The second relates to the differing degree of development of the dorsal portion of the sphenotic spine (character 21 below).

20. Frontal shelf at posterodorsal edge of orbit

In most characiforms with a primitive form of the dilatator fossa, the area at the posterodorsal edge of the orbit has developed a lateral shelf that is continuous with the dorsal area of the sphenotic spine. This shelf can vary from moderately to well developed in these groups.

In all cynodontines, the frontal shelf at the posterodorsal margin of the orbit is either reduced or absent. The reduction of the posterior frontal shelf is probably associated with the type of insertion of the dilatator operculi muscle which, according to Howes (1976: 215), has two origins, one on the dorsal surface of the frontal and a second from the space between the frontal and orbitosphenoid. A reduction of the frontal shelf would provide a passageway for the muscle to reach the ventral portion of the frontal. A reduced frontal shelf at the posterodorsal margin of the orbit also occurs in Acestrocephalus sardina, Acanthocharax microlepis, and Gnathocharax among examined characiform outgroups. However, the dilatator operculi muscle in these species apparently lacks a ventral origin as in cynodontines (Howes, 1976: 235). This region of the frontal in Acestrorhynchus is restructured in a different way, associated with the specialization of the dilatator fossa in this genus (coded as a missing entry for this character).

In Hydrolycus tatauaia and H. armatus (fig. 5), the frontal shelf at the posterodorsal edge of the orbit, although reduced, is present and is slightly more developed in larger specimens. A further reduction of the shelf of the frontal occurs in Cynodon, Rhaphiodon, Hydrolycus wallacei, and H. scomberoides (figs. 4, 6–8). In these taxa the shelf is completely lacking. According to the most parsimonious hypotheses of cynodontine intrarelationships, the lack of the frontal shelf at the posterodorsal edge of the orbit is interpreted as a derived condition, synapomorphic for the Cynodontinae, with a partial secondary reversal to a small shelf in Hydrolycus tatauaia and H. armatus as a derived character for the clade formed by these two species.

21. Dorsal portion of sphenotic spine

Cynodon, Rhaphiodon, and Hydrolycus scomberoides have a reduction in the dorsal portion of the sphenotic spine proximal to the main body of the neurocranium such that the sphenotic spine no longer contacts the ventrolateral margin of the frontal (figs. 4, 7, 8). In all other Hydrolycus species the dorsal portion of the sphenotic spine extends dorsally to contact the margin of the frontal, a condition similar to that found in most characiform outgroups.

A reduction of the dorsal portion of the sphenotic spine is also common to ctenoluciids, erythrinids, and lebiasinines. However, this reduction is accompanied by a longitudinal expansion of the remainder of the spine, causing a restructuring (Vari, 1995: 14). No such restructuring of the spine occurs in the Cynodontinae and the modification in cynodontines does not seem to correspond to that in the groups mentioned above. Characidiins have a reduced sphenotic in which the sphenotic bone excludes the origin of the dilatator operculi muscle from the lateral margin of frontal, a condition not directly comparable to other characiforms (Buckup, 1991: 216). Serrasalmus has a modified sphenotic spine, a feature apparently related to the condition of the levator arcus palatini muscle (Machado-Allison, 1985: 25). Modifications of the sphenotic spine were also described and proposed as a synapomorphy for an assemblage within distichodontids (Vari, 1979: 285–289). Such modifications are different from those described for the Cynodontinae. The reduction of the dorsal portion of the sphenotic spine in cynodontines is hypothesized as synapomorphic for the clade formed by Cynodon and Rhaphiodon with an independent acquisition in Hydrolycus scomberoides.

22. Anterior shelf of frontal

The anterior portion of the frontal shelf in cynodontines also shows some modifications as a consequence of the expansion of the dilatator fossa and the area of attachment of the dilatator operculi muscle.

Hydrolycus armatus has the least modified form of such a shelf among cynodontines relative to the condition in characiform outgroups. Anteriorly the shelf of the frontal in this species covers the dorsal surface of the lateral ethmoids, extending laterally relative to the longitudinal axis of the body. In smaller specimens (LACM 43295-36, 135.9 mm SL; MZUSP 32607, 2 specimens, 151.4–200 mm SL) the anterior portion of the frontal shelf is represented by a very narrow bony platform but is wider in larger specimens. In Hydrolycus tatauaia the anterior shelf of the frontal is similar to that in H. armatus, although relatively less pronounced. Even the largest specimen of H. tatauaia prepared for osteological examination (MZUSP 32632, 257.0 mm SL) has a narrower shelf than smaller specimens of H. armatus. In H. wallacei the anterior shelf of the frontal is completely absent (largest specimen prepared for osteological examination, MZUSP 32638, 163.0 mm SL). As noted in H. armatus, this shelf is considerably more developed in relatively large specimens (examined in dry skeletons preparations, AMNH 91342SD, 450 mm SL). Since Hydrolycus tatauaia reaches sizes comparable to that of H. armatus, it is possible that the shelf also is more developed in larger specimens of the latter species. Hydrolycus wallacei does not seem to reach sizes comparable to the latter two species (largest specimens examined, MCNG 21901, 335 mm SL), but it is possible that specimens larger than those available for study might have the frontal shelf more developed.

Cynodon, Rhaphiodon, and Hydrolycus scomberoides have the anterior shelf of the frontal relatively more expanded than in H. armatus. It also extends more ventrally relative to the longitudinal axis of the body than does the shelf in H. armatus. As a consequence a deep groove is formed between the shelf of the frontal and the orbitosphenoid. In H. scomberoides the anterior frontal shelf is truncated at the midline of the orbit. Although the anterior shelf of the frontal is expanded in all of the three taxa mentioned above, in each of them the shelf has a different shape. In Rhaphiodon the margin of the anterior shelf of the frontal is straight ventrally, whereas in Cynodon it has a curved edge. In Hydrolycus armatus the area between these two elements is shallower and more open. The different conditions of the anterior shelf of the frontal in cynodontines are partially shown in figures 4–8.

A frontal bone with an expanded platform located lateral to the supraorbital canal and projecting over the lateral ethmoid was noted by Buckup (1991: 215) for some characiforms e.g., Charax, Cynopotamus, Tetragonopterus, Bryconops, Oligosarcus and Phenacogaster. That condition resembles that of Hydrolycus armatus within cynodontines and was also noted for Roestes and Gilbertolus. However, the condition of a relatively expanded anterior shelf of the frontal of Cynodon, Rhaphiodon, and Hydrolycus scomberoides is considered derived. However, proposing this feature as synapomorphic for clades within the Cynodontinae is problematic due to the high degree of variation of the shelf in the three taxa, which complicates ordering them in a sequence that diverge from the putatively primitive condition. Furthermore, in Hydrolycus wallacei the shelf is absent, a condition different from that of other cynodontines. Consequently, the various conditions of the anterior shelf of the frontal are herein only hypothesized as autapomorphic for Hydrolycus scomberoides, H. wallacei, and Rhaphiodon vulpinus, with the condition of Cynodon considered synapomorphic for the genus.

23. Third posttemporal fossa bordered by epioccipital and exoccipital

All cynodontines possess a vertically ovate posttemporal fossa bordered by the epioccipital and exoccipital. The hypothesized plesiomorphic condition of the posttemporal openings in characiforms consists of a dorsal and a posterolateral fossa on each side of the neurocranium (Vari, 1979: 289; 1983: 37). The dorsal fossa is bordered by the supraoccipital, the parietal, and the epioccipital. The posterolateral fossa is bordered by the pterotic and epioccipital. A third posttemporal fossa bordered by the epioccipital and exoccipital was described for the Citharinidae and Distichodontidae by Vari (1979: 289), who also discusses the presence of such a fossa in cynodontines. Although a third posttemporal fossa also occurs in the families Curimatidae (Vari, 1983: 37), Hemiodontidae, and Parodontidae (Roberts, 1974), and Hydrocynus (Brewster, 1986) the fossa in those taxa is contained entirely within the epioccipital (Vari, 1983: 37; Brewster, 1986: 168).

In the present study, a third posttemporal fossa similar to that of cynodontines (i.e. bordered by both the epioccipital and exoccipital) was noted in Roestes, Gilbertolus atratoensis, Heterocharax, Lonchogenys, and Gnathocharax sp., and it is hypothesized as a derived condition. Whether or not this feature would provide support for a hypothesis of a close relationship between all the taxa involved depends on a detailed analysis of characiform relationships focusing on a higher level of universality.

24. Dorsal posttemporal fossa

Cynodon species lack the dorsal posttemporal fossa that is bordered by the supraoccipital, parietal, and epioccipital in characiform outgroups. The lack of this fossa, (also in the citharinids Citharinus and Citharidium) was hypothesized as derived by Vari (1979: 289). Carnegiella vesca among the Gasteropelecidae has a reduced dorsal posttemporal fossa (Weitzman, 1954: 218). In the present study such a fossa was found to be apparently lacking in the examined specimens of Gasteropelecus sternicla. The taxa mentioned above are not closely related to cynodontines, with the relationships of Citharinus and Citharidium being with the Distichodontidae (Vari, 1979). The relationships of the Gasteropelecidae are not fully resolved, but the family has been hypothesized as being most closely related to some characid lineage close to Astyanax, Brycon, and Bryconamericus (Weitzman, 1954: 231). Therefore, the lack of the dorsal posttemporal fossa is hypothesized as independent in citharinids, gasteropelecids, and Cynodon with the condition in the latter genus considered synapomorphic.

25. Basioccipital

In all cynodontine species the portion of the basioccipital that articulates with the vertebral column is flared posteriorly, forming a receptacle for the first centrum. This feature was previously noted by Nelson (1949: 501) for Rhaphiodon vulpinus. Ventrally, this posterior projection of the basioccipital is indented and forms two ventral processes. The posterior expansion of the basioccipital becomes even more pronounced in larger specimens in which it covers the first centrum almost completely. The first centrum in these specimens is only seen from a dorsal view, and ventrally from the indented portion of the basioccipital projection. A condition of the craniovertebral joint as described for cynodontines is unique to this assemblage among examined outgroups and hypothesized as a synapomorphy.

Suspensorium and Hyoid Arch

26. Hyomandibula

In all Hydrolycus species and Rhaphiodon vulpinus, the shaft of the hyomandibula that rests against the dorsomedial face of the preopercle is largely separate from the main body of the bone and forms a short process that extends ventrally from the condylar articular surface for the opercle (fig. 9A). In Cynodon, such a process is absent, and the contact of the hyomandibula with the preopercle is in the form of a groove, extending along the arm of the hyomandibula (fig. 9B),—a condition widespread in characiform outgroups. The condition present in Hydrolycus and Rhaphiodon is unique among examined outgroups.

Two alternative, equally parsimonious, hypotheses are possible for this character. The first is the acquisition of the process in the ancestor of the Cynodontinae clade, with its subsequent loss in Cynodon. Alternatively, the process may have arisen independently in the genera Hydrolycus and Rhaphiodon.

27. Symplectic

The symplectic in cynodontines is fairly elongate, its posterior portion bearing a bladelike process extending dorsally from its main body and fitting snugly into a groove in the medial face of the lower arm of the hyomandibula (fig. 10A). This groove is roofed over dorsally. As a consequence, the dorsalmost portion of the symplectic is completely enclosed within the hyomandibula, a condition absent in all other examined characiforms.

In most characiforms the contact of the symplectic with the hyomandibula is formed exclusively by a synchondral joint (fig. 10B) (which is also present in cynodontines) lateral to the dorsal extension of the bone.

Erythrinids also have a well-developed symplectic. However, the condition in this family differs from that in the Cynodontinae. In Hoplerythrinus unitaeniatus and Erythrinus erythrinus the lamellar process of the symplectic extends toward and contacts the metapterygoid (a slight contact of a lamellar process of the symplectic with the metapterygoid was also observed in Boulengerella lateristriga among ctenoluciids). The symplectic process and the hyomandibula are in slight contact along their margins. The symplectic in Hoplias (fig. 10C), in addition to contacting the metapterygoid, extends dorsally to also contact the medial surface of the hyomandibula. Although there is a shallow groove in the portion of the hyomandibula where the dorsal process of the symplectic fits, this groove is never roofed over dorsally as in cynodontines. Erythrinids are more closely related to ctenoluciids, lebiasinids, and Hepsetus (Vari, 1995), taxa which lack an extended contact between the symplectic and the hyomandibula, so that the development of an intimate contact between these two elements in cynodontines and erythrinids is most parsimoniously hypothesized to be an independent acquisition.

Given its unique nature within characiforms, the condition observed in the Cynodontinae is proposed as synapomorphic for the subfamily.

Ontogenetic variation in this character was observed in Rhaphiodon vulpinus. Specimens of 41.6–48.5 mm SL (USNM 231549) have the process on the symplectic extending dorsally. But at these sizes the groove on the hyomandibula, into which the process fits in larger specimens, is absent. In a larger specimen (BMNH 1935.6.4: 33–39, 95.3 mm SL) the groove and the dorsally situated roof associated with it are both present. At this stage the symplectic process fits in the groove of the hyomandibula but does not extend all the way into the roofed portion. In a larger specimen, MZUSP 32812 (191.0 mm SL), the process on the symplectic extends all the way into the roofed portion of the groove in the hyomandibula.

28. Metapterygoid teeth

Metapterygoid teeth are found in all Hydrolycus species and Rhaphiodon vulpinus. Of the two cleared and stained specimens of H. scomberoides examined (MZUSP 26177, 120.6 mm SL, MZUSP 32093, 143.4 mm SL), only the smaller has metapterygoid teeth, which are also present in one dry skeleton of H. scomberoides examined (AMNH 40087, 241.0 mm SL). It is unlikely that this is a function of ontogenetic change of this feature since the intermediate-size specimen (143.4 mm SL, mentioned above) lacks teeth on the metapterygoid. Rather it seems that there is intraspecific variation in the presence of these teeth in H. scomberoides. All examined specimens of Cynodon species lack metapterygoid teeth.

The metapterygoid teeth are found mainly on the dorsal half of the ossification, distributed in patches of different sizes. Some specimens have a broad patch of metapterygoid teeth and in others the patches are very small. The teeth are arranged in separate tooth plates fused with the metapterygoid or can be individually coalesced directly with that bone.

Within actinopterygians, toothed metapterygoids have been reported only for Amia and Polypterus (Arratia and Schultze, 1991) and were described as separate tooth plates that fuse to each other and with the metapterygoid and appear late in ontogeny.

The presence of teeth in the metapterygoid of Hydrolycus species and Rhaphiodon vulpinus is a unique condition in characiforms and there are two equally parsimonious explanations for the distribution of this character. The first is the acquisition of metapterygoid teeth in the ancestor of the Cynodontinae, with its subsequent loss in Cynodon. Alternatively, metapterygoid teeth may have had independent origins in Hydrolycus and Rhaphiodon vulpinus.

29. Mesopterygoid teeth

Cynodontines possess a broad patch of very small teeth that almost entirely cover the buccal surface of the mesopterygoid.

These teeth are present only in a few characiforms: they were observed in a few (but not all) Acestrorhynchus species and in Lebiasina bimaculata (coded as absent by Vari, 1995: 7). Within ctenoluciids, Vari (1995: 24) mentioned the presence of mesopterygoid teeth in Boulengerella lateristriga and B. maculata. Such dentition was also observed in one specimen of B. cuvieri (MZUSP 24162, 138 mm SL). Among erythrinids, such teeth were found in this study in Hoplerythrinus unitaeniatus and Hoplias sp. In the latter species (MZUSP 32372, 86.8 mm SL) there is only a very small patch of teeth on the lower surface of the mesopterygoid of the left side of the head. On the right side there are only two teeth loosely attached to the mesopterygoid. According to Vari (1995: 7) and Oyakawa (personal commun.) mesopterygoid teeth are absent from Hoplias. It is possible that there is interspecific and/or ontogenetic variation in this feature within Hoplias, Boulengerella cuvieri, and Lebiasina bimaculata. Within the Characinae (sensu Eigenmann, 1909), mesopterygoid teeth were observed in one specimen of Roeboides sp. (AMNH 40198, 64.9 mm SL) in the form of a toothplate attached to the mesopterygoid bone. A smaller specimen from the same lot (51.6 mm SL) and one specimen of R. paranensis (MZUSP 19830) lack mesopterygoid teeth. Teeth on the mesopterygoid were absent in all specimens of Roeboides examined by Lucena (1993: 116). In addition, mesopterygoid teeth have also been reported in Crenuchus (Buckup, 1991: 265).

Mesopterygoid teeth are a feature that varies among taxa that have been proposed as closely related to cynodontines: they are absent in Roestes and Gilbertolus, but present in most Acestrorhynchus species. Therefore, the presence of teeth in the mesopterygoid cannot be unequivocally proposed as an additional synapomorphy for the Cynodontinae at this time.

The presence of teeth in other characiform groups mentioned above (except Acestrorhynchus) is most parsimoniously interpreted as an independent acquisition relative to cynodontines.

30. Ectopterygoid teeth

All cynodontines have very small teeth arranged in a patch covering most or all of the medial surface of the ectopterygoid bone (fig. 11).

Teeth on the ectopterygoid occur in diverse characiforms including Acestrorhynchus (Menezes, 1969: 35), the Ctenoluciidae (Vari, 1995: 23), various lebiasinids, the Erythrinidae (Weitzman, 1964: 144–145), Oligosarcus (Menezes, 1969: 12,15), the Serrasalminae (Machado-Allison, 1983: 169, and 1985: 33 for ontogenetic variation), the Characidiinae (Buckup, 1993: 241), Crenuchus (Buckup, 1991: 265), and Xenocharax. Weitzman (1964: 134) suggested that the presence or absence of ectopterygoid teeth may be of little phylogenetic importance, because of considerable intraspecific variation and variation among closely related species.

However, the morphology and pattern of teeth arrangement in the ectopterygoid bone seem to provide some information about relationships in a few characiform groups. Within ctenoluciids, for instance, Vari (1995: 23) reported differences in ectopterygoid teeth patterns between Ctenolucius and Boulengerella and hypothesized that the condition of very small teeth was derived for the latter genus. The serrasalmids Serrasalmus and Pristobrycon have cuspidate ectopterygoid teeth (Machado-Allison, 1985: 33) as opposed to the conical teeth present in the majority of characiforms. These were hypothesized by that author as a synapomorphy for these genera (Oligosarcus was herein observed to also have cuspidate ectopterygoid teeth). Within erythrinids, the size and distribution of teeth on the ectopterygoid bone also differ. No differences in pattern of ectopterygoid teeth were observed in cynodontines. All members of the group have a broad patch of very small teeth covering most of the ventral surface of that bone, a condition similar to that described by Vari (1995: 23) for Boulengerella. In the remaining characiform outgroups in which mesopterygoid teeth are present, they are usually larger relative to the condition observed in cynodontines and Boulengerella. The latter genus is, however, more closely related to Ctenolucius (Vari, 1995) which has a different pattern of ectopterygoid teeth. The condition of the ectopterygoid teeth in Boulengerella and cynodontines is most parsimoniously interpreted as independent acquisitions, with the condition in the latter taxa hypothesized as synapomorphic.

31. Ectopterygoid-metapterygoid contact

In all cynodontines the posterior portion of the ectopterygoid is firmly attached to the anteroventral portion of the metapterygoid. Among characiforms such an ectopterygoid-metapterygoid contact was observed only in Hepsetus. In the examined outgroups the posterior portion of the ectopterygoid contacts only the anterodorsal portion of the quadrate, a condition that also occurs in cynodontines. Hepsetus is more closely related to the Erythrinidae and Ctenoluciidae (Vari, 1995); therefore, this additional contact of the posterior portion of the ectopterygoid to the anteroventral portion of the metapterygoid is most parsimoniously interpreted as an independent acquisition in Hepsetus and the Cynodontinae, with the condition in cynodontines hypothesized as synapomorphic.

32. Ectopterygoid-mesethmoid contact

All cynodontines have longitudinally elongate ectopterygoids extending well beyond the anterior portion of the palatine, and constituting the anteriormost portion of the palatine arch. Longitudinally elongate ectopterygoids are not unique to cynodontines, also occurring in ctenoluciids, Hepsetus and Acestrorhynchus.

The type of contact between the ectopterygoid and the area near the vomer-mesethmoid joint is distinct in Cynodon from that in Hydrolycus and Rhaphiodon. In Cynodon the anterior tip of the ectopterygoid contacts the posterior portion of the mesethmoid just dorsal to the ventral processes on the latter ossification. Although Hydrolycus and Rhaphiodon also have an elongate ectopterygoid, this ossification is in close contact with the ventral surface of the vomer (see character 11) and its anteriormost tip does not contact the mesethmoid in the same way it does in Cynodon.

Typically in characiforms the ectopterygoid is connected to the area near the vomer-mesethmoid articulation by a sheath of connective tissue; however, these elements are not in direct contact, e.g., the Heterocharacini, sensu Géry (1966), Cynopotaminae, Menezes (1976), Roestes, Gilbertolus, Charax, Agoniates, lebiasinids, and gasteropelecids. In all of these taxa the palatine bone constitutes the anteriormost portion of the palatine arch, with the tip of the ectopterygoid lying posterior to it.

In Acestrorhynchus species the ventral surface of the vomer is anteriorly expanded over the ventral surface of the mesethmoid and bears two ventral process to which the palatine and the ectopterygoid abut. The ectopterygoid contacts the vomer along the area of contact between the mesethmoid and vomer. Within erythrinids, Hoplias has a tooth-bearing ossification connected to the anterior end of the ectopterygoid. The anterior tip of this ossification contacts the ventral process of the mesethmoid. However the identity of this ossification is unclear (Weitzman, 1964: 146; Roberts, 1969: 417). In Hepsetus the region occupied by the mesethmoid and vomer is a single ossification and the ectopterygoid inserts into two grooves in the anteroventral surface of this element (Roberts, 1969: 406). In all these taxa the ectopterygoid is not in direct contact with the mesethmoid. Among examined characiforms a direct contact between these two elements was observed in Ichthyborus quadrilineatus and I. besse species more closely related to the Citharinidae-Distichodontidae clade (Vari, 1979). The type of attachment of the ectopterygoid to the vomer and mesethmoid in these Ichthyborus species is unique to those taxa and is of a different type than that in cynodontines. The condition observed in Cynodon is hypothesized as derived for cynodontines and synapomorphic for the genus.

33. Branchiostegal rays

All cynodontines have five branchiostegal rays. Among examined characiforms outgroups, five branchiostegal rays is also present in erythrinids (Weitzman, 1964: 147), the characidiin Characidium fasciatum (Buckup, 1992: 1069), and among hemiodontids in some species of Hemiodus and Bivibranchia (Roberts, 1974: 417, 420). Five branchiostegal rays were also noted for Crenuchus (Buckup, 1991; Vari, 1995), Piaractus nigripinis (Roberts, 1969: 422), and Thoracocharax (Weitzman, 1960). Géry (1962: 271) mentions that Agoniates ladigesi has five branchiostegal rays, of which the anterior two rays are rudimentary. The specimen of Agoniates sp. examined in the present study has, however, only four well developed branchiostegal rays.

The presence of four branchiostegal rays is the most common condition in characiforms. Five branchiostegal rays are present in the groups discussed above. A few groups have only three branchiostegal rays (e.g., pyrrhulinines, Weitzman, 1964: 150; and some anostomids of the genera Pseudanos and Anostomus, Winterbottom, 1980: 39).

A high number of branchiostegals has been considered the primitive condition (McAllister, 1968: 68, 176). All the taxa mentioned above that possess five branchiostegal rays have been hypothesized as being more closely related to characiforms with four branchiostegal rays (see references above). Therefore, at the level of inclusiveness of the present study, the presence of five branchiostegal rays is most parsimoniously interpreted as being derived and having independent origins. The presence of five branchiostegal rays is, as a consequence, proposed as a synapomorphy for the Cynodontinae.

34. Branchiostegal rays on posterior ceratohyal

All cynodontines have two branchiostegal rays on the posterior ceratohyal. This condition is present only in Acestrorhynchus and Ctenolucius (Vari, 1995: 25) among examined outgroups. All other characiforms have only a single ray on the posterior ceratohyal.

Ctenolucius is more closely related to Boulengerella and other taxa with one branchiostegal ray on the posterior ceratohyal (Vari, 1995) and, therefore, the presence of two branchiostegal rays in the posterior ceratohyal is interpreted as derived at the level of inclusiveness of the present study, and hypothesized as being independently acquired in Ctenolucius and cynodontines.

In view of the evidence pointing toward phylogenetic relationships between Acestrorhynchus and the Cynodontinae, the presence of two branchiostegal rays in the posterior ceratohyal cannot be unambiguously proposed as an additional synapomorphy for the Cynodontinae at this time.

Jaws and Dentition

35. Dentary canines

Cynodontines possess a single row of conical teeth in the upper and lower jaws. The teeth are variable in size with small conical teeth alternating with larger canines. One of the anterior dentary canines is enlarged relative to the remaining teeth and extends into the upper jaw when the mouth is closed (see character 36). Enlarged dentary canines are not restricted to cynodontines among characiforms. Relatively pronounced dentary canines also occur in Acestrorhynchus, Hydrocynus, Hepsetus, Ichthyborus, the Heterocharacini, the Erythrinidae, and the Cynopotaminae. In some of these taxa, the canines can get relatively stout, especially in large specimens (e.g., Hydrocynus and Hoplias); however, they are relatively shorter than in cynodontines. In most of these taxa the enlarged canines are not restricted to a single tooth but there are instead a number of relatively large canines of similar size along the dentary and also in the upper jaw (e.g., Hydrocynus, Hoplias, and Galeocharax). In cynodontines one of the anterior dentary canines is always considerably larger than the remaining teeth. A similar canine enlarged to the same degree relative to the remaining comparable teeth is observed in Roestes, Gilbertolus, and Agoniates.

Within cynodontines, a further increase in the relative size of the dentary canine occurs in Hydrolycus tatauaia, H. scomberoides, and H. armatus. In these species the dentary canine is much more developed than in Cynodon, Hydrolycus wallacei, and Rhaphiodon, in which only the dorsalmost tip of the dentary canine extends dorsally to reach the snout opening. However, in Hydrolycus tatauaia, H. scomberoides, and H. armatus it extends far dorsally into the snout opening. The highly developed dentary canines of H. tatauaia, H. scomberoides, and H. armatus represent a unique condition within characiforms synapomorphic for this assemblage.

36. Foramen in anterior portion of snout for dentary canine

The highly developed dentary canine in cynodontines extends dorsally into the upper jaw. The anterior portion of the snout has been restructured, resulting in an opening through which the dentary canine fits when the mouth is closed. A large opening in the anterior portion of the snout is delimited anteriorly and laterally by the premaxilla, posteriorly and posteromedially by the ascending process of the maxilla, and anteromedially by the vomer-mesethmoid (fig. 11). This opening lies just ventral to the nasal openings in the neurocranium, and the tips of the highly enlarged dentary canine of Hydrolycus tatauaia, H. scomberoides, and H. armatus extend through the nasal openings when the mouth is closed.

The arrangement of the bones in the anterior part of the snout in characiform outgroups with enlarged canines differs from that described for cynodontines. In Roestes, Gilbertolus, and Agoniates the enlarged dentary canine extends dorsally to fit in a space delimited anteriorly and laterally by the premaxilla and the ascending process of the maxilla, and medially by the palatine. The most pronounced difference from the condition observed in cynodontines is that the ascending process of the maxilla passes posterior to the dentary canine in cynodontines while in Roestes, Gilbertolus, and Agoniates this process lies anterior to the dentary canine. The ascending process of the maxilla in those three taxa rests on the posterior surface of the premaxilla, the typical condition in characiforms (e.g., Brycon, Weitzman, 1962: 32). In cynodontines the ascending process of the maxilla has shifted posteriorly and is not in close association with the premaxilla, leaving an intervening space through which the dentary canine passes.

In Acestrorhynchus (Menezes, 1969) and Hepsetus the upper jaw has a different modification to receive the canine teeth from the dentary. These two genera have relatively elongate premaxillae, which usually bear two foramina in Acestrorhynchus and one in Hepsetus through which the dentary canines passes when the mouth is closed. Hydrocynus has wide spacing of teeth in both jaws (Brewster, 1986: 187) with the teeth on the upper jaw being intercalated by teeth from the lower jaw when the mouth is closed. The enlarged teeth on the anterior portion of the snout lie outside the mouth when it is closed, and they fit into shallow grooves between the teeth of the opposing jaw.

The condition of the foramen for the dentary canine in the anterior portion of the snout, with the ascending process of the maxilla shifted posteriorly, not contacting the premaxilla, and forming the posterior and lateromedial portion of the foramen, is unique for cynodontines among examined characiforms and hypothesized as synapomorphic.

37. Replacement tooth trenches

There is considerable variation in the morphology of the replacement tooth trenches in characiforms. Monod (1950) and Roberts (1967) described different types of replacement tooth trenches in the order. According to Roberts, cynodontines, Hepsetus, Salminus, and the Ctenoluciidae have “shallow replacement trenches at the base of the functional tooth rows, and the replacement teeth lie right at the base of the functional teeth and are readily observed.” The majority of other characiforms were described as having more deeply excavated replacement tooth trenches with the replacement teeth lying considerably below the bases of the corresponding functional teeth (Weitzman, 1962: 33; Roberts, 1967: 233). In cynodontines, all teeth in the mandibular replacement trench are horizontally aligned, with the tip of the replacement teeth projecting posteriorly. In the upper jaw, including the premaxilla and maxilla, there is no conspicuous trench, but the replacement teeth are also horizontally and posteriorly directed. In the majority of examined characiform outgroups, including Roestes, Gilbertolus, Acestrorhynchus, the Heterocharacini, and Charax, the replacement teeth are placed more deeply in the trench and angled vertically to slightly posterodorsally relative to the functional teeth.

Horizontally and posteriorly directed teeth in the replacement trench also occur in Hepsetus (Monod, 1950; Brewster, 1986: 187) and Hydrocynus (Roberts, 1967: 233; Brewster, 1986: 187). The morphology of the replacement tooth trench in the latter genus differs from that of Hepsetus and the Cynodontinae (Brewster, 1986: 170–171). In the Ctenoluciidae the replacement teeth are also positioned at a 90° angle relative to the functional teeth, in a horizontal relative to the longitudinal axis of the body, and not vertically or slightly angled as in the majority of characiform outgroups. The replacement teeth are very small with their tips somewhat posteriorly directed. Hepsetus and ctenoluciids are hypothesized as being more closely related to erythrinids and lebiasinids (Vari, 1995), families with replacement tooth trenches more similar to that described for the majority of characiforms. The similar conditions shared by Hepsetus and cynodontines are most parsimoniously hypothesized as independent acquisitions, with the cynodontine condition considered synapomorphic.

Branchial Arches

38. Gill-rakers

All gill-rakers on the leading portion of the first ceratobranchial in cynodontines consist of small, flat bony plates with very small spines covering their entire lateral surface (fig. 12A, B). This condition is present only in Acestrorhynchus (Menezes, 1969) among examined outgroups.

Within characiforms a variety of different forms of gill-rakers occurs along the length of the first ceratobranchial, in addition to that described above for cynodontines and Acestrorhynchus. In the majority of characiforms the first ceratobranchial has conical, elongate gill-rakers with or without teeth. Erythrinids, ctenoluciids, and Hepsetus (Roberts, 1969: 423) and genera assigned to the Cynopotaminae (Menezes, 1976) have elongate gill-rakers along most of the first ceratobranchial, with a few anterior reduced rakers, similar in shape to those described for cynodontines and Acestrorhynchus.

Roberts (1969: 423) considered the condition present in Rhaphiodon and Acestrorhynchus as primitive for characiforms based on the similarity of the gill-rakers in these groups and some primitive actinopterygians (e.g., Polypterus, Lepisosteus, and Esox). However, given the widespread occurrence of elongate gill-rakers along the first ceratobranchial in different characiform lineages, the first ceratobranchial covered with spiny gill-rakers over its entire length in cynodontines and Acestrorhynchus is most parsimoniously interpreted as being derived at the level of inclusiveness of the present analysis.

In view of the evidence pointing toward phylogenetic relationships between Acestrorhynchus and the Cynodontinae the presence of spiny gill-rakers cannot be unambiguously proposed as an additional synapomorphy for the Cynodontinae at this time.

39. Spines on gill-rakers

The differences in the spines on the free dorsal margin of the gill-raker within the Cynodontinae are noteworthy. Enlarged spines on the free dorsal margin of the gill-rakers of the first ceratobranchial were observed in Cynodon and Hydrolycus wallacei (fig. 12A). These spines are considerably larger than the remaining spines. Usually one of the spines is distinctly larger, with its length more than half of the vertical length of the bony plate of the gill-raker. Enlarged spines are more evident on the gill-rakers situated toward the posterior end of the first ceratobranchial. Hydrolycus tatauaia, H. scomberoides, H. armatus, and Rhaphiodon share a different condition (fig. 12B) in which the spines are not very developed relative to the remaining spines on the surface of the gillraker, being only slightly larger than the latter.

Acestrorhynchus species, the only other characiform taxa that have short, flattened gill-rakers covering the entire first ceratobranchial, have considerably enlarged spines on the free upper edge of the gill-rakers. In all examined species except A. falcirostris and A. nasutus there are one or more prominent spines larger than the others. The latter two species lack a single, more prominent spine on the dorsal margin of the gill-rakers, with all spines on the free upper edge of the gill-raker being rather similar in size (Menezes, 1969: 61,74–75). Nevertheless, the spines on the free upper edge of the gill-raker in these species are also considerably larger than the ones on the surface of the gill-rakers. In view of the current hypothesis of relationships between Acestrorhynchus and cynodontines, the presence of relatively enlarged spines on the dorsal margin of the gill-rakers of the latter species could be proposed as the primitive condition. Ontogenetic information provides additional support for this hypothesis. In juveniles of both Cynodon (AMNH 32485SW, 35 mm SL), Rhaphiodon (USNM 231549 41.7–50.1 mm SL), and Hydrolycus (MCNG 3204, 25.6–38.6 mm SL) the spines on the free upper edge of the gill-raker are conspicuously enlarged. In larger specimens of Rhaphiodon they become smaller relative to the remaining spines on the surface. Therefore, the condition observed in juveniles, provides information about the direction of the character transformation, with the condition observed later in their development, i.e., spines on the free edge of the gill-raker not conspicuously enlarged relative to those on its surface, considered derived.

This character is most parsimoniously hypothesized as synapomorphic for the clade formed by Hydrolycus tatauaia, H. scomberoides, and H. armatus with an independent origin in Rhaphiodon.

40. First hypobranchial

The anterior portion of the first hypobranchial in cynodontines is anteroventrally prolonged into a prong-shaped process that extends from the ventrolateral margin of the main body of the bone (fig. 13). The anterior articular cartilaginous surface of the first hypobranchial is situated on the tip of this process.

In examined characiform outgroups, the first hypobranchial is either a flattened (e.g., Erythrinidae, Lebiasinidae, Roestes, Gilbertolus, Gnathocharax, Charax, and Brycon) or an elongate ossification (e.g., Acestrorhynchus, ctenoluciids, and Hepsetus) that varies from straight to slightly slanted anteroventrally, but which typically lacks the conspicuous elaboration noted above for cynodontines. Such unelaborated ossifications were also observed among African characiforms (e.g., Xenocharax and Distichodus fasciolatus), hypothesized as primitive members of the order (Fink and Fink, 1981). Unelaborated hypobranchials have been hypothesized as plesiomorphic (Vari, 1983: 11). Certain anostomids (Laemolyta taeniata, Anostomus anostomus), prochilondontids (Prochilodus), and Serrasalmus have some kind of anterior elaboration on the first hypobranchials. However, none of these groups shows the degree of elaboration presented by cynodontines. In all these groups, the anterior process of the first hypobranchial is much shorter and less curved than in cynodontines. In Serrasalmus, in addition, the process is dorsoventrally flattened with no associated articular surface.

The condition observed in cynodontines is unique within characiforms and hypothesized as synapomorphic.

41. First ceratobranchial

The first ceratobranchial in cynodontines has its anterior portion curved dorsally, forming a distinct angle relative to the longitudinal axis of the remaining portion of that ossification. In Hydrolycus and Rhaphiodon, the angle and the extension of this projection is only slightly pronounced, but it is clearly distinct from the condition observed in characiform outgroups in which the first ceratobrachial is straight over its entire length. A condition similar to that observed in Hydrolycus and Rhaphiodon was observed only in the cynopotamine genus Galeocharax.

The dorsally directed projection of the first ceratobranchial is particularly well developed in Cynodon (fig. 13), in which genus the projection is about twice as long as in Hydrolycus and Rhaphiodon and the angle of the projection relative to the remaining portion of that ossification is much more pronounced. A straight, elongate first ceratobranchial with no anterior projection is considered a primitive condition for characiforms (Vari, 1983: 11). Galeocharax is hypothesized to be more closely related to taxa with straight first ceratobranchials (e.g., Acestrocephalus, and Charax; Menezes, 1976, Lucena, 1993) and, therefore, the dorsally directed first ceratobranchial in Galeocharax is most parsimoniously hypothesized as an independent acquisition from that in cynodontines.

The dorsally directed first ceratobranchial of cynodontines forming an angle relative to the longitudinal axis of the remaining portion of that ossification is considered a synapomorphy for cynodontines with its further enlargement in Cynodon considered a synapomorphy for the genus.

Anterior Vertebrae

42. First vertebra

Cynodon species possess two processes on the ventral portion of the first centrum that lie just dorsal to the ventral processes of the basioccipital (see character 25), and are directed posteriorly and slightly laterally. These processes were observed in all examined cleared and stained specimens and dry skeletons of Cynodon species, including a juvenile (AMNH 32485, 35.0 mm SL) in which two small bony projections are present on the ventral surface of the first centrum.

In one specimen of Hydrolycus wallacei (MZUSP 32638, 152.2 mm SL; this feature could not be examined in the other available specimen) two small bumps of bone are present in the same region of the ventral processes as in Cynodon but are much less developed. Weakly developed processes were also observed in larger specimens of Hydrolycus armatus (AMNH 91342, 450 mm SL), but were lacking in one specimen of Hydrolycus sp. (AMNH 40048, 117.7 mm HL). Ventral processes on the first centrum are lacking in the examined specimens of H. tatauaia, H. scomberoides, and Rhaphiodon vulpinus.

Ventral processes on the first centrum are lacking in all examined characiform outgroups and their presence in cynodontines is considered a derived feature. In the present study only well-developed processes on the ventral surface of the first centrum in Cynodon species will be considered due to the lack of more specimens of Hydrolycus wallacei and Hydrolycus tatauaia for examination. Additional specimens could show that the presence of bony projections on the ventral portion of the first centrum is more widespread among cynodontines. However, the relatively more highly developed condition of these processes which is unique to Cynodon, can be considered a synapomorphy for this genus.

43. Neural complex of Weberian apparatus

In all cynodontine species the neural complex of the Weberian apparatus does not directly contact the posterior surface of the neurocranium. In the primitive condition among characiphysans (Fink and Fink, 1981: 324) the neural complex is tilted anteriorly and articulates with the posterior margin of the neurocranium. Among cynodontines the neural complex has a more vertical orientation relative to the longitudinal axis of the body, resulting in a gap between its anterior margin and the posterior margin of the neurocranium. At the anterior edge of the neural complex there is a short process onto which attach ligaments arising from the posterior edges of the supraoccipital crest and exoccipital.

The lack of contact of the neural complex with the posterior portion of the neurocranium in cynodontines is unique among examined outgroups, and is considered a synapomorphy for the group.

In all cynodontines the transverse processes of the second and third vertebrae contact each other in an interdigitating pattern that forms an immovable articulation, a condition unique within examined characiforms. Such an interdigitating articulation is primarily the consequence of two modifications of the transverse processes of the second and third vertebrae described below (characters 44 and 46).

44. “Transverse process” of second vertebra

The second vertebra of cynodontines bears a lateral process designated as the “transverse process” by Nelson (1949). This process is highly developed and slightly posterodorsal oriented. It is bifurcated distally into two short processes that clasp the transverse process of the third neural arch (figs. 14–16).

The presence of a “transverse process” on the second centrum is not unique to cynodontines among examined characiforms. Such a process also occurs in erythrinids, ctenoluciids, Hepsetus, Heterocharacini, Gnathocharax, Roestes, Gilbertolus, and in some Acestrorhynchus species. However, in all of these taxa the process is only slightly developed and is represented only by a small prominence of bone situated ventral to the intercalarium and the transverse process of the third neural arch. There are ligamentous attachments between the latter process and the “transverse process” of the second vertebra. In erythrinids the process is enlarged relative to that in the remaining examined outgroups and although it contacts the transverse process of third neural arch, it does not bifurcate distally as it does in cynodontines.

The modification of the “transverse process” of the second vertebra in cynodontines is a unique condition among examined characiforms and hypothesized as a synapomorphy for that clade.

45. Lateral process of second vertebra

The lateral process of the second vertebra is highly modified in cynodontines. In the usual characiform condition the process is an elongate element extending laterally and somewhat anteriorly with no modifications of its distal portions (see Weitzman, 1962: 36 for Brycon meeki).

Within cynodontines a number of differences from the generalized characiform condition are observed. Cynodon and Hydrolycus species (figs. 14 and 15, respectively) (except Hydrolycus scomberoides, see further comments on the condition of this species below) have the lateral process shortened laterally and greatly expanded dorsoventrally. It extends slightly anteriorly proximally and then turns posteriorly distally. Ventrally, at its tip, there is a broad surface for the attachment of ligamentous tissue connecting with the pectoral girdle. The posterolateral margin of the process ends in two short processes. Ligamentous tissue attaches to the posterior margin between the two processes and extends to the lateral process of the fourth vertebra. The anteromedial portion of the lateral process of the second vertebra is in contact with the posterior margin of the basioccipital.

Hydrolycus scomberoides has a slightly different condition from that described above. In this species the lateral process of the second vertebra is dorsoventrally flattened and posteriorly directed. The distal tip of the process has a broad surface for the attachment of ligamentous tissue connecting with the pectoral girdle as described for Cynodon and the remaining Hydrolycus species. However, instead of being directed ventrally, this broad surface in H. scomberoides is directed laterally. Anteriorly the lateral process contacts the posterior margin of the basioccipital. The posterolateral margin is a simple structure without processes that serves as an area of attachment for the ligamentous tissue from the process of the fourth vertebra.

The lateral process of the second vertebra is highly modified in Rhaphiodon vulpinus (fig. 16) relative to both the primitive characiform condition and the condition in the remaining cynodontines. The condition in R. vulpinus has been previously briefly described by Weitzman (1962: 36) and was illustrated by Nelson (1949: 517). In a dorsal view there is a triangular-shaped flat sheet of bone extending laterally and posteriorly under the tripus. Ventrally a short process extends anteriorly, and overlapping the ventral portion of the first centrum. A longer process extends posteriorly and slightly laterally under the ventral portion of the centrum of the third vertebra.

The highly modified lateral process of the second vertebra in cynodontines is a unique condition among characiforms and is interpreted as synapomorphic for the subfamily with the conditions in H. scomberoides and Rhaphiodon vulpinus constituting autapomorphies for these species.

46. Lateral process of third vertebra

All cynodontine species except Hydrolycus wallacei possess a lateral process on the third vertebra in addition to the typical characiform transverse process of the third neural arch. This additional process abuts the lateral portion of the “transverse process” of the second vertebra that extends posteriorly to contact the third vertebra. In Rhaphiodon, Hydrolycus tatauaia, and H. armatus, the process is relatively well developed, originating just ventral to the lateral portion of the “transverse process” of the second vertebra, and is more evident in larger specimens. A slightly different condition occurs in Cynodon and Hydrolycus scomberoides. In these species the process is relatively shorter than in Rhaphiodon, Hydrolycus tatauaia, and H. armatus and apparently originates posteriorly to the portion of the “transverse process” of the second vertebra that contacts the transverse process of the third neural arch (fig. 14), rather than ventral to that process as in Rhaphiodon, Hydrolycus tatauaia, and H. armatus (figs. 15 and 16). This process is lacking as a separate element in the two examined cleared and stained specimens of Hydrolycus wallacei. Among examined characiform outgroups only Ctenolucius has a process similar to that described for Rhaphiodon, Hydrolycus tatauaia, and H. armatus.

The explanations for the distribution of this character are equivocal under any type of optimization used. It is my opinion that this ambiguity is partly a consequence of the poor understanding of the anatomy of this process. A better understanding of this feature would come from a more detailed examination of this element by means of finer dissections at the region of the third vertebra, examination of different ontogenetic stages of this features in specimens having the two different conditions described above, and examination of additional specimens of Hydrolycus wallacei to confirm the absence of the process in this species. It was not possible to carry out a more detailed study of this character at this time. Therefore, the additional lateral process on the third vertebra will not be hypothesized as a derived feature at any level for the present analysis.

The contact between the “transverse process” of the second vertebra with the processes of the third vertebra described under characters 44 and 46 forms the interdigitating articulation mentioned above. In large specimens of Hydrolycus armatus (e.g., AMNH 40048, 117.7 mm HL) an additional process originates at the posteroventral portion of the second vertebra and extends posteriorly to abut the lateral process of the third vertebra, contributing an additional element to this pattern of the articulation.

The transverse process of the third neural arch is shorter in cynodontines than in examined characiform outgroups. Some variation in the shape is observed within the group, with Hydrolycus species (especially H. armatus and H. wallacei) having a slightly more elongate process. Variation in this feature among cynodontines will not be taken into further consideration at this time because of the difficulty of unambiguously establishing discrete units to account for this variation.

47. Tripus

The tripus of Rhaphiodon vulpinus is anteroposteriorly elongate (fig. 16). The blade portion of the tripus is particularly elongate anteriorly, with its tip almost reaching the posterior of the neurocranium. The tripus in the remaining cynodontine species is similar to that of characiform outgroups in which the blade portion of the tripus does not (or only slightly) extend anteriorly of the main body of the tripus. An anteroposteriorly elongate tripus was described for ctenoluciids by (Vari, 1995: 27), in which, the elongation is accomplished by a continuity between the anterior portion of the ossification and the posterior section that terminates in the transformator process. The anterior bladelike portion of the tripus is not elongate in ctenoluciids as it is in Rhaphiodon vulpinus. In the latter species the posterior section that terminates in the transformator process is not continuous with the anterior portion of the tripus (see character 48 below).

The anteriorly elongate blade portion of the tripus in Rhaphiodon vulpinus is unique among examined characiforms and hypothesized as autapomorphic for the species.

48. Transformator process of tripus

The posterior section of the tripus that terminates in the transformator process is distinct in Rhaphiodon vulpinus from that in other cynodontines and among characiform outgroups. In the typical characiform condition the transformator process of the tripus is a thin curved process continuous with the posterior portion of the tripus with no modifications medially (Fink and Fink, 1981: 331). In Rhaphiodon vulpinus the posterior portion of the tripus is almost horizontally aligned. In the region medial to the lateral portion of the fused fourth pleural rib plus parapophysis the tripus becomes much thinner and is medially directed at an approximately 90° angle with the anterior portion of that ossification. The transformator process of the tripus in Rhaphiodon vulpinus ends in an enlarged rectangular bony plate (figured by Nelson, 1949: fig. 5F, labeled as “T”). The structure of the transformator process in Rhaphiodon vulpinus is unique among examined characiforms and considered autapomorphic for the species.

49. Fourth pleural rib + parapophysis

The fused fourth pleural rib + parapophysis in Cynodon and Rhaphiodon differs notably from that of Hydrolycus. In the latter genus the fused pleural rib and parapophysis extends laterally in a flat, thin dorsoventrally oriented process (fig. 15). This condition is similar to the condition observed in characiform outgroups and is described and figured by Weitzman (1962: 36, 68).

In Cynodon and Rhaphiodon the process is greatly enlarged, extending anteriorly and ventrally, and covering the posterolateral portion of the tripus (figs. 14 and 16). The distal end of the process is greatly enlarged, forming a broad articular surface for the attachment of ligamentous tissue. Such a process was previously described for Rhaphiodon vulpinus by Nelson (1949: 500), who termed the enlarged distal portion of the process a basal plate. This basal plate contacts the medial portion of the pectoral girdle in the region where the cleithrum articulates with the supracleithrum. A lateral process of the fourth vertebra as described for Cynodon and Rhaphiodon is a unique condition within examined characiforms and hypothesized as derived.

A heavy mass of connective and ligamentous tissue stretches from the distal portions of the processes of the second and fourth vertebrae, and extends to the pectoral girdle in the region of the articulation of the cleithrum and supracleithrum. An understanding of connections between the elements of the Weberian apparatus and the pectoral girdle in cynodontines necessitates a much more detailed study of these connections in order to isolate the different ligaments and their origins and insertions, a task not feasible at this time.

Lesiuk and Lindsey (1978) in their study on head bending in Rhaphiodon vulpinus mentioned that the attachment of the Weberian apparatus to the dorsal tip of the cleithrum constrains the movement of the latter element. The cleithrum is not rigidly attached to the bones of the Weberian apparatus, allowing a rotation of the girdle with respect to the skull through an arc of approximately 10°.

The highly modified fused fourth pleural rib and parapohysis of Cynodon and Rhaphiodon is unique among characiforms and is hypothesized as synapomorphic for these two genera.

50. Dorsal articulation between fourth and fifth vertebrae

All cynodontines have an articulation between the fourth and fifth vertebra that is unique among examined characiforms. The posteroventral portion of the fourth vertebra in cynodontines has a process extending posterolaterally, with ligamentous attachments originating from both its dorsal and ventral surfaces. Ligaments from the dorsal surface of the process of the fourth vertebra attach to a process extending anteriorly and laterally from the dorsal portion of the fifth centrum (figs. 14–16). Ligaments from the ventral surface of the process of the fourth vertebra attach to the anteroventral portion of the fifth vertebra, which also bears another process (see character 51 below).

Although modifications of the fourth and fifth vertebrae to form an articulation is found in some examined outgroups, the condition differs from that in cynodontines. For instance, Agoniates, Triportheus, and Brycon have an expansion of the posteroventral portion of fourth centra extending to the anteroventral portion of the fifth vertebra. The contact with the fifth vertebra is only from the dorsal surface of the process of the fourth vertebra. The process in these taxa is only a thin lamella of bone that extends posteriorly from the vertebra, a condition different from that in cynodontines in which the process is enlarged and has a lateral extension. There is no modification of the fifth vertebra. Caenotropus has a process originating at the posterior portion of the fourth vertebra that contacts a groove on the anterior portion of the fifth vertebra. Serrasalmus has a process similar to that of cynodontines (enlarged with an dorsal articular surface) but originating at the posterior region of the fourth centrum, in a more dorsal position than that in cynodontines. This process contacts a process originating from the neural arch of the fifth vertebra.

The articulation between the fourth and fifth vertebra in cynodontines is apparently related to the ability of these fishes to rotate the head upward, as described by Lesiuk and Lindsey (1978) in Rhaphiodon vulpinus. These authors observed that upward rotation of the head in this species occurs at the articulation between the fourth and fifth vertebrae (indicated by arrows in their figures 1C, 1D and 1E). Upward rotation of the head is yet to be studied in the remaining cynodontine species. Although the articulation between the fourth and fifth vertebra in Rhaphiodon vulpinus is more developed than that of the remaining cynodontine species (see character 51 below), the presence of such an articulation in the remaining species may imply that they may demonstrate at least some degree of head rotation relative to the vertebral column.

The type of articulation between the fourth and fifth vertebra in cynodontines is unique among characiforms and hypothesized as synapomorphic.

51. Ventral articulation between fourth and fifth vertebrae

In addition to the process at the dorsal portion of the fifth centrum mentioned above, cynodontines possess another process situated at the anteroventral portion of the fifth vertebra which is, in turn, laterally expanded, extending ventral to the lateral process of the fourth centrum that forms the articulation between the fourth and fifth vertebrae (figs. 14–16). The ligamentous tissue from the ventral surface of the process of the fourth vertebra attaches to this lateral process of the ventral portion of the fifth vertebra. The latter process is more evident in larger specimens, in which it is more pronounced laterally and extends more ventrally to contact the process of the fourth vertebra. In Hydrolycus scomberoides the process seems to be less pronounced than in the remaining cynodontines, hardly apparent in a 143.4 mm SL specimen (MZUSP 32093), and small in a 241.0 mm SL specimen (AMNH 40087). In specimens of the other cynodontine species within the size range of those of H. scomberoides, the process of the fifth vertebra is more developed. It is possible to see that there is a slight lateral expansion on the fifth vertebra relative to the same portion of the sixth vertebra which lacks such a process. Therefore, this feature is regarded as present in H. scomberoides.

In Rhaphiodon vulpinus the ventral process of the fifth vertebra is further enlarged relative to that of other cynodontines, being as well-developed as the other process of the fifth vertebra that articulates dorsally with the process of the fourth vertebra. The process is more conspicuous when observed either from a ventral and/or lateral view (fig. 16). It is already evident in a 50.1 mm SL specimen (USNM 231549), and very conspicuous in a 96.8 mm SL specimen (BMNH 1935.6.4: 33–39).

The presence of the ventral process of the fifth vertebra is unique to cynodontines among characiforms and hypothesized as a synapomorphy for this taxa, with its further enlargement in Rhaphiodon vulpinus hypothesized as autapomorphic for this species.

52. Pleural ribs

The fifth and to a certain extent the sixth and seventh pleural ribs of Rhaphiodon vulpinus are modified relative to that of remaining cynodontines and examined characiform outgroups. The fifth pleural rib in that species is very short relative to the more posterior ribs and has a flattened proximal portion and a very slender distal portion. A process arises from the medial portion of the rib and extends medially in the region ventral to the fifth centrum. Another process arises at the dorsal tip of the rib and extends posteriorly to the sixth vertebra (fig. 16). These features of the fifth rib are present also in the sixth and seventh ribs, although they are not as conspicuous and show some variation in different specimens. The more posterior ribs are similar to those of other cynodontines and characiform outgroups.

The fifth rib in the remaining cynodontine species is slightly shorter than the more posterior ribs. It is slender along its entire length and possesses a very short medially directed process dorsally, which is less developed than that in Rhaphiodon. The posterodorsal process present in Rhaphiodon is absent from the remaining cynodontines. A process on the medial portion of the fifth rib (similar to those of Cynodon and Hydrolycus) also occurs in some characiform outgroups (e.g., Gilbertolus, Roestes, Acestrorhynchus, Agoniates, Gnathocharax, and the Heterocharacini), and was described and figured by Roberts (1969: 426, figs. 46, 47); see also Buckup (1991: 226), and Lucena (1993: 56) for distribution of this feature in characiforms. In these taxa, however, the rib is not conspicuously flat in its proximal half as in Rhaphiodon and is approximately the same length as the more posterior ribs.

The structure of the fifth rib of Rhaphiodon vulpinus is unique among examined characiforms and regarded as autapomorphic for this species.

53. Parapophyses

The parapophyses of all precaudal vertebrae posterior to the fifth vertebra in cynodontines are longitudinally elongate, with the parapophysis of one vertebra extending anteriorly and articulating with the vertebra anterior to it. The first enlarged parapophysis is that of the sixth vertebra (figs. 14, 15), gradually becoming less pronounced in vertebrae posterior to it.

The parapophyses of the vertebrae of other examined characiforms are also somewhat elongate, having a process extending from the portion of the parapophysis that articulates with the vertebral centrum. The condition observed in cynodontines seems to be the result of an enlargement of that process of the parapophysis associated with articulation at a more ventral portion of the centrum. In outgroups the lateral fossae with which the parapophyses of the anterior precaudal vertebrae articulate have a more central position on the centrum. The process extending from each parapophysis is oriented slightly anteroventrally, reaching the anteroventral portion of the vertebra but not extending beyond it to reach the anterior vertebra. The fossae of the posterior precaudal vertebrae gradually shift to a more ventral position on the lateral portion of the centrum with the process of the parapophysis also decreasing in size. In cynodontines the lateral fossae with which the parapophyses of the anterior precaudal vertebrae articulates are in a more ventral position on the centrum than that in examined outgroups and the process of the parapophysis is oriented anteriorly and not ventrally.

In addition to cynodontines an articulation between the parapophysis of one vertebra to the vertebra anterior to it was observed only in Hydrocynus among examined characiform outgroups. Hydrocynus is hypothesized as being the sister group to certain Alestes (Brewster, 1986: 192), a genus lacking an articulation between the parapophysis of one vertebra and the vertebra anterior to it as described for cynodontines and Hydrocynus. As a consequence, this feature is hypothesized as an independent acquisition in Hydrocynus and the Cynodontinae and considered synapomorphic for the latter taxa.

54. Baudelot's ligament

In all cynodontine species (except Rhaphiodon vulpinus) Baudelot's ligament is strong and attaches to the ventral portion of the enlarged lateral processes of the second vertebra (fig. 15). This constitutes a third point of attachment of this ligament in addition to the typical characiform attachment anteriorly to the basioccipital and the posteriorly to the pectoral girdle. Rhaphiodon has a somewhat different condition from the remaining cynodontines. In Rhaphiodon Baudelot's ligament is not as well developed and although it contacts the lateral process of the second vertebra ventromedially, it is not attached to that process as it is in the remaining cynodontines. The condition in Rhaphiodon largely resembles that of characiform outgroups.

Two alternative, equally parsimonious, hypotheses are possible for this character. The attachment of Baudelot's ligament with the lateral process of the second vertebra may be hypothesized as synapomorphic for the Cynodontinae, with a secondary loss of the attachment in Rhaphiodon vulpinus, or as independently acquired in Cynodon and Hydrolycus.

Pectoral Girdle

55. Postcleithrum 2

A lack of postcleithrum 2 is common to all cynodontines. Among characiforms a lack of this ossification also occurs in Gilbertolus, Gnathocharax (Lucena, 1993), Ctenolucius, Boulengerella lateristriga, B. maculata (Vari, 1995: 26), Hepsetus (Roberts, 1969: 426; Vari, 1995: 27), and the Gasteropelecidae (Weitzman, 1954: 226).

Three postcleithra along the posterior margin of the pectoral girdle are plesiomorphic for characiforms (Roberts, 1969: 426; Vari, 1983: 35; 1995: 26), and the lack of postcleithrum 2 is considered a derived character. The reduction in the number of postcleithra occurs also within other ostariophysan lineages including some gonorhynchiforms, cypriniforms, and siluriforms and it was interpreted (see Fink and Fink, 1981: 334) as specializations that evolved independently in these lineages.

In view of the current evidence for a close relationship between the Cynodontinae and Gilbertolus plus Roestes, the lack of postcleithrum 2 might represent a derived condition for these taxa (with a reversal in Roestes). The lack of this element could also be alternatively considered an independent loss in Gilbertolus and the Cynodontinae.

56. Postcleithrum 3

Cynodontines also lack postcleithrum 3, a feature that is shared with various other characiforms: Gilbertolus, Roestes, Gnathocharax, ctenoluciids (Vari, 1995: 26), Iguanodectes (Lucena, 1993: 115), Hepsetus (Roberts, 1969: 426), gasteropelecids (Weitzman, 1954: 226), and Chilodus (Vari et al., 1995: 9).

This feature was hypothesized as a synapomorphy for the clade formed by cynodontines, Gilbertolus, and Roestes by Lucena and Menezes (1998).

57. Coracoid

The posterodorsal portion of the coracoid in all cynodontine species is perforated by a very large foramen (figs. 17, 18) situated just ventral to the insertion of the radials of the pectoral fin and the articulation with the mesocoracoid. The presence of a posterodorsal foramen in the coracoid was previously noted by Starks (1930: 169) and Nelson (1949: 507) for Rhaphiodon vulpinus. A condition similar to cynodontines was observed only in Gilbertolus among examined outgroups.

In some characiform outgroups (e.g., Roestes, Acestrorhynchus, Heterocharax, Agoniates, Boulengerella, Hydrocynus, Hemiodus, Brycon, Oligosarcus, and Triportheus) a foramen of various sizes is present at the region ventral to the articulation with the mesocoracoid and sometimes extending posteriorly ventral to the insertion of the radials of the pectoral fin. Although present in the characiforms mentioned above, in none of these groups is the posterodorsal foramen of the coracoid as enlarged as in the Cynodontinae and Gilbertolus. Therefore, the condition in these taxa is hypothesized as derived. In Roestes the posterodorsal foramen of the coracoid is intermediate in size between that of Gilbertolus and cynodontines, and that of the remaining characiforms in which it is present. However, there is also variation in the size of the foramen in the remaining characiform outgroups, rendering the definition of discrete states to account for this variation problematic.

Lucena and Menezes (1998) also noted the presence of the enlarged coracoid foramen in cynodontines and Gilbertolus; however, their interpretation of this character was ambiguous because of its absence in Roestes.

58. Size of coracoid

Cynodontines possess relatively enlarged coracoids, a feature shared with a number of other characiforms such as the Gasteropelecidae, Triportheus, Agoniates, and Gnathocharax. In addition, Weitzman (1954: 230) and Roberts (1969: 426) mentioned the presence of enlarged coracoids, in Pseudocorynopoma and Clupeacharax, respectively. The shape of the expanded coracoids differs in all the examined characiforms outgroups that possess them. They are more developed in the Gasteropelecidae and Triportheus than in the Cynodontinae. In the Gasteropelecidae they are fan-shaped with a rounded ventral profile (Weitzman, 1954); whereas in Triportheus and Gnathocharax the expansion occurs mainly in the posterior portion of the ossification. In cynodontines the enlargement of the coracoid occurs at the vertical plane of the longitudinal axis of the body, with the ventral profile of the coracoid being straight (fig. 17) as in characiform outgroups in which this ossification is not enlarged. Agoniates has an expanded coracoid similar in shape to that of cynodontines.

All the taxa mentioned above differ significantly in the rest of their osteology and it seems that expanded coracoids originated independently in all of them, as hypothesized by Weitzman (1954: 228–231). More recently Castro and Vari (1990) provided a detailed discussion about the presence of enlarged coracoids in characiforms and the appropriateness of using that feature to propose relationships. These authors also proposed an independent origin of expanded coracoids in most of these groups, based on the evidence indicating closer relationships of these taxa to species or species groups without expanded coracoids. Taxa proposed as closely related to the Cynodontinae (Roestes, Gilbertolus, Acestrorhynchus) do not have expanded coracoids. The presence of this feature in cynodontines is hypothesized as synapomorphic for the group.

59 Fusion of coracoids

In all cynodontines the coracoids are closely applied to each other along their entire midline. In Cynodon species and Rhaphiodon vulpinus the contralateral coracoids are fused even more along a large portion of the area of contact, with the region of the fusion between the two ossifications densely corrugated. All attempts to separate the left and right elements resulted in breaking of one coracoid in the region of the corrugated area. Fusion of the contralateral coracoids and the corrugated pattern is present in juveniles of Rhaphiodon vulpinus (USNM 231549, 50 mm SL), but not evident in a 35 mm SL specimen of Cynodon (AMNH 32485). In Rhaphiodon vulpinus a triangular-shaped area in the anterior portion of the conjoined coracoid is formed only by one coracoid. This region is located ventral to the dorsal margin of the ossification that forms the ventral edge of the large foramen formed by the cleithrum and coracoid. This feature occurs in all examined osteological preparations of Rhaphiodon vulpinus and was previously noted by Nelson, 1949: 507.

Although all Hydrolycus species have the two portions of the coracoids in very close contact and very difficult to separate, they are not fused to one another, at least not to the degree observed in Cynodon and Rhaphiodon. It was possible to separate completely the opposite coracoids in all specimens in which this was attempted. In some very large specimens of Hydrolycus a small area in the anteriormost portion of the coracoids seemed fused, and one of the sides was damaged in the area of the fusion when the bones were separated. In other large specimens, however, the two coracoids were completely separate and intact. A slight corrugated pattern was observed in one specimen of H. scomberoides, but the contralateral coracoids were not fused.

Within characiforms, fused and corrugated coracoids have been reported only in the Gasteropelecidae (Weitzman, 1954), a taxon not closely related to the Cynodontinae, as discussed above.

The condition of partially fused and corrugated coracoids described for Rhaphiodon and Cynodon is hypothesized as a synapomorphy for these genera.

60. Mesocoracoid

The mesocoracoid in cynodontines is greatly enlarged relative to the condition in the majority of examined outgroups. The typical condition of the mesocoracoid in characiforms is that of a short ossification attached to the cleithrum, scapula, and coracoid (e.g., Brycon, Weitzman, 1962: 41, 76). Within cynodontines the mesocoracoid occupies a larger portion in the medial portion of the pectoral girdle and has broader articular surfaces with the surrounding bones (fig. 17). It is attached to the cleithrum along a vertical crest at the medial portion of this latter ossification, with the dorsal tip of the mesocoracoid extending dorsally to at least half the length of the vertical portion of the cleithrum. In the typical characiform condition the dorsal tip of the mesocoracoid does not extend to the midpoint of the vertical portion of the cleithrum. The articulation of the mesocoracoid with the coracoid in cynodontines extends more posteriorly on this latter ossification compared to the typical characiform condition, extending along the posterodorsal surface of the coracoid just dorsal to the insertion of the pectoral-fin radials. An enlarged mesocoracoid similar to that of cynodontines occurs only in Gilbertolus among examined outgroups. In Agoniates and Gnathocharax the articulation of the mesocoracoid with the cleithrum is somewhat broad with the dorsal tip of the mesocoracoid reaching the midpoint of the vertical portion of the cleithrum. The articulation with the coracoid, however, is similar to the typical characiform condition. Therefore, the overall enlargement of the mesocoracoid is not comparable to that in the Cynodontinae and Gilbertolus.

Lucena and Menezes (1998) also noted the enlarged mesocoracoid in cynodontines and Gilbertolus. However, as with character 58, above, the interpretation of the distribution of this feature was ambiguous considering its absence in Roestes and given the hypothesis of monophyly of Roestes plus Gilbertolus they proposed.

61. Scapula

The scapular foramen in the Cynodontinae is almost entirely exposed when examined from lateral view (fig. 18). In the majority of examined characiform outgroups the scapular foramen is covered laterally by the cleithrum and not exposed in lateral view. The condition observed in cynodontines seems to be the result of a posterior shift of the articulation of the scapula and mesocoracoid relative to the medial surface of the cleithrum. In cynodontines the enlarged mesocoracoid articulates with the cleithrum on the middle of the medial surface of this ossification and the articulation of the scapula is with the posteroventral portion of the medial surface of the cleithrum. In the typical characiform condition the articulation of the mesocoracoid and scapula is more anterior on the medial surface of the cleithrum (Weitzman, 1962: fig. 19; Roberts, 1969: figs. 48–52) and as a consequence the scapular foramen is completely covered by the posterior portion of the cleithrum. The insertion of the pectoral girdle is also shifted posteriorly as a result of the posterior shift of the articulation of the scapula with the cleithrum, so that a vertical through the base of the unbranched pectoral-fin ray is situated posterior to a vertical through the posteriormost margin of the cleithrum.

A posterior shift of the articulation of the scapula and mesocoracoid with the consequent exposure of the scapular foramen and posterior shifting of the articulation of the pectoral girdle occurs also in Gilbertolus among examined outgroups (Lucena and Menezes, 1998). In the Gasteropelecidae the scapular foramen is also exposed from lateral view. However, in that family the pectoral girdle shows a high degree of modification (Weitzman, 1954: 225), including changes in the overall shape of the cleithrum. Thus, the condition in gasteropelecids on the one hand, and that in cynodontines and Gilbertolus on the other, seems to be the consequence of different modifications of the pectoral girdle. In addition in the Gasteropelecidae, the base of the unbranched pectoral-fin ray is located at the vertical through the posterior margin of the cleithrum and not posterior to it, as in the Cynodontinae and Gilbertolus. This feature also has an ambiguous interpretation following the same arguments mentioned above for characters 57 and 60.

62. Cleithrum

The anterior portion of the cleithrum in cynodontines ends in a vertically elongate process that articulates with the anterior portion of the enlarged coracoids (fig. 18). This feature is also present in characiform outgroups with enlarged coracoids such as Triportheus and Agoniates, but is absent from the Gasteropelecidae. In the remaining examined characiform outgroups the anterior portion of the cleithrum that contacts the anterior portion of the coracoid is continuous with the posterior part of the cleithrum along the entire vertical margin of the latter ossification. Lucena's (1993) study on the family Characidae provided some support for a hypothesis of the relationships of Agoniates. As discussed previously under Historical Overview and Comments on Cynodontine Relationships with other characiforms, additional studies are needed in order to test that hypothesis, but in view of the larger number of derived features shared by cynodontines and other characiforms (e.g., Gilbertolus, Lucena and Menezes, 1998), the features shared by Agoniates and cynodontines are most parsimoniously interpreted as homoplastic. The relationships of Triportheus are not fully resolved, but evidence discussed by Castro and Vari (1990) points toward a possible relationship between Triportheus and some lineage within Brycon, a group that does not possess a cleithrum ending in a vertically elongate process. Therefore, the condition in cynodontines is, hypothesized as synapomorphic.

Within cynodontines, the condition in Rhaphiodon vulpinus differs from that of Cynodon and Hydrolycus. In the latter two genera the anterodorsal tip of the cleithrum that contacts the coracoid is pointed anteriorly and in some cases slightly upturned. This anterodorsal projection is lacking in Rhaphiodon vulpinus in which the anterior portion of the cleithrum ends in a continuous curve. In the present analysis the condition of the anterior portion of the cleithrum in Rhaphiodon vulpinus is most parsimoniously interpreted as autapomorphic for the species.

The cleithrum shows some differences among cynodontines, the most pronounced occurring at the posteroventral margin. In Rhaphiodon vulpinus the posteroventral margin of the cleithrum is fused to and continuous with the scapula. This species lacks a free margin continuous along the entire posterior and ventral margins of the cleithrum.

In all Hydrolycus and Cynodon species the entire posterior and ventral margins of the cleithrum are continuous, with the posteroventral margin of the cleithrum not being fused to and continuous with the scapula. The condition in Hydrolycus and Cynodon is widespread for characiforms, with the condition in Rhaphiodon occurring only in gasteropelecids, among examined outgroups, and hypothesized as independently acquired.

The margin of the cleithrum dorsal to the level of insertion of the pectoral fin is slightly indented in Cynodon and Hydrolycus species, with the portion of the bone just anterior to the insertion of the unbranched pectoral-fin ray slightly expanded. The expanded portion is pointed in H. scomberoides, whereas in Cynodon and in the remaining Hydrolycus species, this portion of the bone is rounded. However, the establishment of unambiguously discrete states for this feature was complicated because of variation of this portion of the ossification in the examined cynodontines. As a consequence this feature is not proposed as an additional character of phylogenetic interest.

Pelvic Fins

63. Pelvic fin insertion

The pelvic fin in Hydrolycus scomberoides is distinctly dorsal of the ventral profile of the abdomen. As a result the ventral profile of the abdomen continues without interruptions to the anus. In other cynodontines the pelvic fin is inserted at the ventral profile of the abdomen, which is thus interrupted by the pelvic-fin insertion, the typical condition among characiforms.

In serrasalmines the pelvic fins are inserted slightly dorsal to the ventral profile of the body, this being more evident in Mylossoma sp. The condition in serrasalmines is associated with the modification of the scales at the ventral profile of the body resulting in a serrated ventral keel. The pelvic fin inserts at the dorsal portion of the modified scale of the ventral keel, a condition different from that of Hydrolycus scomberoides.

The pelvic fin inserted dorsal to the ventral profile of the abdomen in Hydrolycus scomberoides is hypothesized as autapomorphic for this species.

Anal Fin

64. Anal-fin rays

Cynodon species have 61–80 branched anal-fin rays. All other cynodontines combined have 27–50 branched anal-fin rays. The high number of branched anal-fin rays in Cynodon is unusual among characiforms. The clade formed by ctenoluciids, erythrinids, lebiasinids, and Hepsetus has very short anal fins with 13 or fewer rays. The majority of other characiforms have up to 38 branched anal-fin rays (see Lucena, 1993: 61, 118 for the distribution of this feature among characiform taxa). Characiforms with relatively more branched anal-fin rays include Stethaprion crenatum (36–42; Reis, 1989: 57); Piabucus caudomaculatus (36–38; Vari, 1977: 3), and Charax (33–56, Lucena, 1987); Roestes and Gilbertolus (38–47 and 40–48 respectively, Menezes and Lucena, 1998). The presence of 60 or more branched anal-fin rays is unique to Cynodon among characiforms and hypothesized as derived.

Caudal Fin

65. Hypurals

Cynodon and Hydrolycus species have hypurals 1–3 fused into a single unit that is articulated with the PU¹ + U¹ (fig. 19). In Rhaphiodon only hypurals 2 and 3 are fused into a single unit that is articulated with PU¹ + U^1. Although the dorsal margin of hypural 1 is in very close contact with the unit formed by the fusion of hypurals 2 and 3, it still remains a separate ossification. Although not articulated with the PU¹+ U¹, the anterior end of hypural 1 extends in a thin process directed toward PU¹ + U¹. In one specimen of Rhaphiodon vulpinus (LACM 43295-64, 138.4 mm SL) hypural 1 is articulated with PU¹ + U¹, and it is somewhat fused to hypurals 2 and 3. In the largest examined cleared and stained specimen of Rhaphiodon vulpinus (MZUSP 32812, 191 mm SL) hypural 1 was free from the unit formed by hypurals 2 and 3.

In all juvenile cynodontine specimens available for osteological examination Rhaphiodon vulpinus: (USNM 231549, 41.7–50.1 mm SL) and Cynodon sp. (AMNH 32485SW, 35 mm SL) all six hypurals consist of separate elements. Hypurals 2 and 3 are still not fused in one specimen of Rhaphiodon vulpinus (BMNH 1935.6.4: 33–39, 96.8 mm SL). Hypurals 1–3 are already fused in a specimen of Cynodon gibbus (LACM 43295-89, 102.4 mm SL).

Fusion of the three ventral hypurals seems to occur in at least some Acestrorhynchus species. One cleared and stained specimen of A. falcatus (AMNH 43418, 140.0 mm SL) has the anterior portions of hypurals 1–3 fused. The posterior portions of these elements also seem fused, although the suture between the bony plates is still evident. In a smaller specimen of the same species (MZUSP 4572-91, 111.1 mm SL) the three ventral hypurals can still be distinguished as distinct elements, albeit in very close contact along their proximate margins. One dry skeleton of A. heterolepis (AMNH 93088) has the three ventral hypurals completely fused. One specimen of A. microlepis (AMNH 40106, 80 mm SL) has hypurals 1 and 2 fused into a single element. One specimen of A. lacustris has hypurals 1 and 2 fused anteriorly. Hypurals 1–3 of A. nasutus and A. falcirostris are separate. The condition for Acestrorhychus was coded as a missing entry.

Other characiforms that show fusion between hypurals are the Citharinidae-Distichodontidae assemblage, Hemiodontidae, Serrasalminae, Crenuchus, and Poecilocharax, discussed by Vari, 1979: 313). In all these taxa, the fusion is only of hypurals 1 and 2.

Roestes and Gilbertolus possess separate hypurals, however the presence of some degree of fusion in hypurals of Acestrorhynchus, a genus hypothesized as closely related to the Cynodontinae, renders the interpretation of this character ambiguous.

Intermuscular Bones

66. Epineurals

In all cynodontines the medial branch of one of the anterior epineural bones contacts the lateral surface of the neural complex of the Weberian apparatus.

An account of details of the intermuscular bones in teleosts was given by Patterson and Johnson (1995). Among ostariophysans, they recorded the intermusculars of the characiform Alestes dentex and three cypriniforms and pointed out the necessity of a fuller survey of the anterior intermusculars in otophysans. A complete survey of this character system is beyond the scope of the present study; however, examination of cynodontines and outgroups showed that there is a large degree of variation of intermuscular bones in characiforms. The condition present in cynodontines is, however, unique to this group within examined characiforms and hypothesized as synapomorphic.

67. Myorhabdoi

Rhaphiodon vulpinus possesses highly developed myorhabdoi (sensu Chapman, 1944) characterized by long, slender bones, branched dorsally, that project anteroventrally along the sides of the body dorsal to its longitudinal midline. Myorhabdoi are also present along the dorsal-fin pteryogiophores. The myorhabdoi are especially developed in the anterodorsal portion of the body, where they are almost horizontally aligned. In this region, some of the myorhabdoi bundle into a single, thick bone that attaches to the posterior surface of the neurocranium. Howes (1976: 224) and Lesiuk and Lindsey (1978) described this bundle of intermuscular bones as being attached to the pterotic. Examination of this feature shows, however, that this tendon attaches to the region of the supraoccipital just posterior to the posterior margin of the parietal, and dorsal to the epioccipital.

Patterson and Johnson (1995) mentioned the presence of myorhabdoi in various teleost groups as autapomorphic for those taxa in which they occur. Within characiforms, myorhabdoi are present in the gasteropelecids (Weitzman, 1954: 224), and in Citharinus. In none of these taxa are the myorhabdoi as developed as in Rhaphiodon. In gasteropelecids some of the intermuscular bones also attach to the neurocranium, at the region of the supraoccipital crest (Weitzman, 1954: 218). These are apparently epipleurals (labeled as epl in figure 2 in Weitzman, 1954: 244). There is also an intermuscular bone attached to the posttemporal + supracleithrum (figured but not labeled in figure 2 in Weitzman, 1954: 244). I was unable to determine whether it consists of a myorhabdoi. In Citharinus the myorhabdoi do not attach to the neurocranium.

Although closest relatives remain to be elucidated, gasteropelecids do not seem to be closely related to the Cynodontinae, a hypothesis also supported by Weitzman (1954: 230), who, in addition to and Castro and Vari (1990), suggested that gasteropelecid relationships may lie within some tetragonopterine lineage. Citharinus is related to the Citharinidae-Distichodontidae lineage, the other members of which do not possess myorhabdoi. Therefore, in agreement with Patterson and Johnson (1995), the myorhabdoi should be hypothesized as independent acquisitions in the characiform taxa in which they are present.

The presence of myorhabdoi is unique to Rhaphiodon within cynodontines and hypothesized as autapomorphic. Lesiuk and Lindsey (1978) considered that the epaxial musculature pulling on this bundle of intermuscular bones (described by them as forming a cable-like tendon, and labeled as T in their fig. 2), provided the force to rotate the head upward.

Miscellaneous

68. Scales

The scales of Hydrolycus scomberoides are characterized by serrations on their exposed portion (as opposed to cycloid scales in the remaining cynodontine species). This kind of scale was termed spinoid by Roberts (1993: 62). He restricted the term “ctenoid scales,” usually employed for any kind of serrated scales, to scales in which the spines along the margin are formed as separate ossifications. The presence of spinoid scales is not unique to H. scomberoides, among characiforms and the distribution of such scales were discussed by Vari (1995: 28). According to Vari, the various characiforms that possess spinoid scales are more closely related to diverse taxa with cycloid scales. In addition, serrations on the scales of most of the taxa in which they are present differ in form and distribution. The spinoid scales of H. scomberoides most resemble those of Galeocharax gulo illustrated in Roberts (1993: 69, fig. 6). Therefore, spinoid scales are hypothesized as having independent origins in taxa in which they occur and are autapomorphic for H. scomberoides.

69. Gasbladder

The gasbladder of Rhaphiodon vulpinus was described in detail and figured by Nelson (1949) and Lesiuk and Lindsey (1978). The most distinctive feature of the gasbladder in this species is the presence of a series of fringelike appendices along the length of the lateral surface of the posterior chamber. These appendices extend laterally, penetrating into the body wall. Nelson (1949) suggested that the appendices form a refinement for the reception of vibrations and/or pressure changes in the overall functioning of the gasbladder and Weberian apparatus in hearing and hydrostatic perception. Lesiuk and Lindsey (1978) noted the proximity between the appendices in the swim bladder to the pored scales on the lateral-line.

Cynodon and Hydrolycus species lack appendices in the gasbladder. The condition in Rhaphiodon vulpinus is unique among characiforms and is considered autapomorphic.

Monophyly of Cynodontinae

The hypothesis of the monophyly of the Cynodontinae is supported by a series of 24 synapomorphies listed below and discussed in the previous section. Nineteen of these represent characters unique, unreversed, and not further modified within the Cynodontinae. Characters 51 and 62 are proposed as additional synapomorphies for the Cynodontinae with a further modification in Rhaphiodon vulpinus (see monophyly of the latter genus); character 45 is further modified in Rhaphiodon vulpinus and Hydrolycus scomberoides; character 20 is partially reversed in Hydrolycus armatus and H. tatauaia; character 41 is further modified in Cynodon. Characters 26, 28, and 54 are ambiguous.

Synapomorphies:

Ambiguous features for this clade are:

A₁. Process on shaft of hyomandibula, absent in Cynodon (character 26)

A₂. Metapterygoid teeth, absent in Cynodon (character 28).

A₃. Attachment of Baudelot's ligament to ventral portion of enlarged lateral process of second vertebra, absent in Rhaphiodon (character 54).

The Cynodontinae consists of two primary monophyletic lineages, one formed by the genus Hydrolycus and the other by a Cynodon and Rhaphiodon clade.

Monophyly of HYDROLYCUS

Only one uniquely derived feature supports the monophyly of Hydrolycus (character 4). In addition three derived features shared between Hydrolycus species and Rhaphiodon vulpinus (characters 26, 28, and 54) are ambiguous, having two equally parsimonious explanations for their distribution on the most-parsimonious cladogram.

1. Dorsoventral enlargement of mesethmoid spine being almost round in shape from a lateral view, further enlarged in Hydrolycus scomberoides, H. armatus and H. tatauaia (character 4)

Ambiguous features for this clade are:

A₁. Process on shaft of hyomandibula, also present in Rhaphiodon (character 26).

A₂. Metapterygoid teeth, also present in Rhaphiodon (character 28).

A₃. Attachment of Baudelot's ligament to ventral portion of enlarged lateral process of second vertebra, also present in Cynodon (character 54).

Intrageneric Relationships in HYDROLYCUS

Within Hydrolycus, H. wallacei is hypothesized as the sister group to a clade formed by H. scomberoides, H. armatus, and H. tatauaia, which is diagnosed by the following synapomorphies:

Hydrolycus armatus and H. tatauaia share the following synapomorphy:

No unique autapomorphies were discovered for either Hydrolycus armatus or H. tatauaia.

Hydrolycus scomberoides is characterized by the following autapomorphies:

Hydrolycus wallacei is characterized by the following autapomorphies:

Monophyly of CYNODON and RHAPHIODON clade

Monophyly of the clade formed by Cynodon and Rhaphiodon is supported by six synapomorphies:

Monophyly of CYNODON

The two species of Cynodon share nine synapomorphies plus three characters whose distribution can be explained by two equally parsimonious hypotheses:

Ambiguous features for this clade are:

A₁. Lack of a process on shaft of hyomandibula (character 26).

A₂. Lack of metapterygoid teeth (character 28).

A₃. Attachment of Baudelot's ligament to ventral portion of enlarged lateral process of second vertebra, (character 54).

No unique autapomorphies were discovered for either Cynodon gibbus or C. septenarius.

Monophyly of RHAPHIODON

Rhaphiodon is monotypic, represented by a highly specialized species R. vulpinus. The autapomorphies characterizing Rhaphiodon vulpinus are as follows:

Ambiguous features for this species:

A₁. Process on a shaft of hyomandibula, also present in Hydrolycus (character 26).

A₂. Metapterygoid teeth present, also present in Hydrolycus (character 28).

A₃. Lack of attachment of Baudelot's ligament to lateral process of second vertebra (character 54).

SUBFAMILY CYNODONTINAE Eigenmann, 1907

Diagnosis: The subfamily Cynodontinae is diagnosed within characiforms by a series of derived features listed in Monophyly of the Cynodontinae under Phylogenetic Reconstruction above. Externally, cynodontines are most readily distinguished from all other characiforms by their oblique mouth and a pair of highly developed dentary canines.

Remarks: Howes's (1976) concept of the Cynodontini included Roestes. Herein, the Cynodontinae includes only the genera Cynodon, Rhaphiodon, and Hydrolycus. Lucena and Menezes (1998) placed the Cynodontinae (as herein defined) together with the Roestinae (which included Roestes and Gilbertolus) in the family Cynodontidae.

Comments on CYNODON MEIONACTIS

I examined one paratype (MNHN 1998-400) and one nontype specimen of C. meionactis (MNHN 1998-1769). Dr. M. Jégu checked characters of the holotype (MNHN 1998-0397). I concluded that C. meionactis Géry et al., 1999, is different from C. septenarius.

Cynodon septenarius and C. meionactis have a relatively large orbital diameter compared to C. gibbus. Cynodon septenarius can be distinguished from C. meionactis by the lack of the band of dark pigmentation on the base of the caudal-fin rays (fig. 24), present in C. meionactis (Géry et al. 1999: 71, figs. 1 and 3); and in having i,7 pelvic-fin rays compared to i,8 in C. meionactis. The number of anal-fin rays showed a large degree of variation in C. septenarius (61–77) overlapping with that of C. meionactis (which has 63–67, Géry et al., 1999: 70). However, the variation observed in C. septenarius does not demonstrate any geographical pattern.

Géry et al. (1999: 71, 77) also suggested the existence of an undescribed Cynodon species in the upper Rio Negro, however, they did not formally describe it, although the diagnostic features provided by those authors for that form conform to those of C. septenarius. In addition, according to the present study, C. septenarius is the only Cynodon species that occurs in the upper Rio Negro, the area from which the specimens examined by Géry et al. (1999) originated.

Géry et al. (1999) hypothesized that C. meionactis, was endemic to the upper Maroni River basin pending examination of specimens that might originate from Suriname. No Cynodon specimens from Suriname were examined in the present study and the closest known occurrence of the genus in the Guianas is in Guyana where C. gibbus occurs in the Rupununi area and C. septenarius occurs in the lower portions of the Essequibo River drainage and in the Demerara River. Studies of other characiform taxa have revealed species restricted to the Maroni River system of Suriname and French Guiana with some of them also occurring in the Mana River drainage (e.g., Cyphocharax punctatus, Vari, 1992a; Hemiodus huraulti, Langeani, 1996; Semaprochilodus varii, Castro, 1988, also see Planquette et al., 1996).

In the present study, complete descriptions are provided for C. gibbus and C. septenarius. Detailed information on C. meionactis is provided by Géry et al. (1999).