Phylogenetic relationships of the enigmatic Carlastyanax aurocaudatus (Eigenmann) with remarks on the phylogeny of the Stevardiinae (Teleostei: Characidae)

The Stevardiinae is one of the most diverse subfamilies of Characidae, with more than 200 described species. This subfamily comprises the genera included in the “clade A” of Malabarba & Weitzman (2003), defined by having the combination of eight branched dorsal-fin rays and four teeth on the posterior premaxillary row [including the genera traditionally grouped into the Glandulocaudinae by sharing a secondary sexual caudal-fin gland (Eigenmann, 1914; Géry, 1977; Weitzman & Menezes, 1998)] and Aulixidens Böhlke, Bryconadenos Weitzman, Menezes, Evers & Burns, Lepidocharax Ferreira, Menezes & Quagio-Grassioto, Nantis Mirande, Aguilera & Azpelicueta, and Phallobrycon Menezes, Ferreira & Netto-Ferreira (Mirande, 2010; Ferreira et al. 2011). The monophyly of a clade assignable to the Stevardiinae was obtained in all the phylogenies of the Characidae published, Phylogenetic relationships of the enigmatic Carlastyanax aurocaudatus (Eigenmann) with remarks on the phylogeny of the Stevardiinae (Teleostei: Characidae)

The monotypic genus Carlastyanax Géry was created to include Astyanax aurocaudatus Eigenmann (Géry, 1972) (Fig.  1). This species was described in Astyanax by Eigenmann (1913), in spite of the presence of four teeth in the posterior premaxillary row (vs. five in Astyanax). Several diagnostic morphological features, especially the presence of a hooked third dentary tooth, were used by Géry (1972) to justify the creation of Carlastyanax, although some of these features are either present in many species of the Characidae, or are easily explainable as autapomorphies. According to Géry (1972), the characters diagnosing Carlastyanax are: (1) two rows of premaxillary teeth, with anterior row bearing 4-5 tricuspid teeth and posterior row bearing 4, rarely 5, tri-to pentacuspid teeth; (2) anterior nostril tubuliform; (3) branchiostegal membrane not fused to the isthmus, but on its counterpart across the isthmus; (4) gill rakers few, short, and rounded or lanceolate; (5) reduction of the infraorbitals 1, 5, and 6; (6) dorsal fin equidistant between the snout and the end of the hypural; (7) adipose fin present; (8) caudal fin not covered by scales; (9) pelvic fin (i)i, 5 rays; and (10) lateral line almost completely pored, except for the last one to three scales (Géry, 1972). Román Valencia & Ruiz-C. (2005) and Ruiz-C. & Román- Valencia (2006) synonymized Carlastyanax with Astyanax based on the variable form of the hooked third dentary tooth in Carlastyanax, which they compared with variations in the form of this tooth in species of Astyanax, and by the inconsistency, in their opinion, of some additional data presented by Géry (1972) to diagnose Carlastyanax. Ruiz-C. & Román Valencia (2006) cited Bertaco & Malabarba (2001) as providing information on some species of Astyanax with a hooked third dentary tooth, but all dentaries illustrated in their paper have regular posteriorly curved teeth. Furthermore, and more important than the discussion of that character as diagnostic for Carlastyanax, Ruiz-C. & Román-Valencia (2006) synonymized Carlastyanax with Astyanax without considering two issues when a phylogeny is not available to support taxonomic decisions: does Carlastyanax aurocaudatus have the character combination defining a species of Astyanax, or does Carlastyanax aurocaudatus have a character combination suggesting a relationship with a particular group within the Characidae?
Both questions had answers at that time. Carlastyanax aurocaudatus has only four teeth on the posterior premaxillary row (vs. five in Astyanax) and eight branched dorsal-fin rays (vs. nine in most species of Astyanax). Those two features exclude Carlastyanax from Astyanax, regardless of the validity of a hooked third dentary tooth as a "generic" character. However, a relatively well-corroborated phylogenetic analysis of the Characidae was not available when Ruiz-C. & Román- Valencia (2006) synonymized those two genera. Knowledge about characid phylogeny has been much improved in the last years (Mirande, 2009(Mirande, , 2010Mirande et al., 2011), which allow us to test if that synonymy is justifiable, and also to propose a more precise hypothesis about the relationships of Carlastyanax. The goal of this paper is to elucidate the phylogenetic relationships of Carlastyanax in a familial context, especially to test if this genus is valid and if it is related to Astyanax and/or to the Stevardiinae. We also present the most comprehensive phylogeny of the Stevardiinae to date, with 41 species examined and coded for the phylogenetic characters of Mirande (2010) plus 51 species of Creagrutus Günther that were not examined, but coded from observations of Vari & Harold (2001) and Torres-Mejia & Vari (2005).

Material and Methods
Two datasets were analyzed. In the first one, which we refer to as Examined Matrix, 17 species were added to the matrix of Mirande (2009Mirande ( , 2010 and Mirande et al. (2011). First approximations to the analysis herein presented suggested that Carlastyanax was closely related to Creagrutus. In order to test more adequately the relationships between these two genera, a second dataset based on the Examined Matrix, with the addition of 51 species of Creagrutus coded exclusively from literature (Vari & Harold, 2001;Torres-Mejia & Vari, 2005), was generated. This second dataset is herein referred to as the Extended Matrix.
Characters analyzed are those from Mirande (2010) with the addition of one character from Mirande et al. (2011) and 26 characters mostly from Vari & Harold (2001), which are listed below. Although some of these characters are at present uninformative, they were maintained in the analysis considering future potential addition of terminals. A total of 392 characters were analyzed. All phylogenetic analyses were performed with TNT (Goloboff et al., 2008a) under implied weighting (Goloboff, 1993) following the criteria by Mirande (2009Mirande ( , 2010. The implied weighting method takes into account the congruence between characters during tree searches. As the phylogeny is judged to be the most important source of correlation between characters, those characters that are more congruent to each other are also supposed to be the most correlated with the phylogeny, and therefore more weight is attributable to them during searches. This method has a constant K that permit to adjust how much the characters are down-weighted in relation to their homoplasy. As equal weighting is equivalent to using a value of K tending to infinite, it can be considered a particular case of implied weighting (in which the information from character congruence is completely ignored during searches) (Mirande, 2009). Implied weighting was also defended in terms of stability of the analyses to the addition of new information (Goloboff et al., 2008b).
In characters with many missing entries the test of the phylogenetic analysis for the hypotheses of homology defined in each character state is weaker relative to characters with fewer missing data. This usually produce characters with many missing data to have artificially lower number of steps -and hence higher weights -when optimized on most parsimonious trees. The extended implied weighting method, recently implemented in TNT under the command "xpiwe" was used as a solution of that potential problem (Goloboff, 2013). This method prevents the characters with many missing entries to have artificially high weights. This is done for each character by assuming extra steps among the missing entries that are proportional to the homoplasy of the observed distribution of states. This calculation is mathematically equivalent to using different values of K for each character depending on its proportion of missing entries. Characters with many missing data are assigned by this method with lower values of K and consequently the homoplasy is more strongly down-weighted relative to characters with fewer missing entries (for details read Goloboff, 2013). Stability was used as a meta-criterion to choose some values of K to produce the final hypothesis. As in Mirande (2009Mirande ( , 2010 and Mirande et al. (2011) the measures of stability used were the distortion coefficient ("tcomp") (Farris, 1989) and the SPR distance ("sprdiff") (Goloboff, 2008). Clade support was evaluated through Symmetric Resampling, showing values of groups supported/contradicted (GC) (Goloboff et al., 2003) obtained from 200 replicated matrices. Each resampled matrix was analyzed using Sectorial Searches and Tree-Drifting (Goloboff, 1999). Stability of clades is expressed as GC values using as source the consensus of the most parsimonious trees at each value of K, as in Mirande (2009Mirande ( , 2010. All searches were made through a TNT script, which is provided in Supplementary File 1. Specimens were cleared and stained (C&S) following Taylor & van Dyke (1985). Examined specimens, additional to those cited by Mirande (2010) and Mirande et al. (2011)

Results
Character Descriptions. Twenty-six characters are herein added to the list of 366 characters presented by Mirande (2010) and Mirande et al. (2011). Character 123 from Mirande (2009Mirande ( , 2010 is similar to characters 368 and 369 in this study. Therefore that character and those herein added are described in some detail starting from the number 367. The complete list of characters is provided in the Supplementary File 2. Most of the characters added were coded following the definitions of Vari & Harold (2001). References to their numbers in the phylogeny of Vari & Harold (2001) are given after the string "VH" in the corresponding character definitions. The data matrix is presented in the Supplementary File 3, but the coding of the examined species relevant for the phylogeny of Carlastyanax is also discussed.
The anterior row of premaxillary teeth is usually composed of aligned teeth or it has one or two teeth whose bases are slightly more anterior than the remaining ones (state 0). In some species the anterior teeth at the premaxilla, which have intraosseous development, are not aligned and form two recognizable rows at least in some portion of the bone (state 1). In this state even the developing teeth that are lodged inside the premaxilla are not aligned, and the alveolar portion of the premaxilla is consequently expanded longitudinally (i.e., anteroposteriorly) relative to the species with only one row of intraosseous premaxillary teeth. This character is similar to characters 368 and 369 (see below).
In most species of the Characidae, the third dentary tooth is rather similar to the remaining teeth (state 0). The hooked form of the third dentary tooth was used by Géry (1972) to justify the creation of Carlastyanax, including A. aurocaudatus as its single species. In this state the third dentary tooth is strongly decurved posteriorly relative to the remaining teeth from the dentary (state 1; Fig. 2). This character was observed only in Carlastyanax aurocaudatus, and it constitutes an autapomorphy of this terminal. 368. Premaxillary dentition: (0) one or more series of similar teeth; (1) a recognizable triad of larger teeth with rounded base.
In most characids there is one or two series of premaxillary teeth, which are rather homogeneous in form and size or gradually decrease in size laterally (state 0). Vari & Harold (2001) recognized a different dentition in most species of Creagrutus and Piabina Reinhardt, genera whose species have a medial triad of larger teeth with rounded base. This triad is usually margined anteriorly by teeth from a primary premaxillary row of teeth, according to Vari & Harold (2001) Fig. 3). The homology of the teeth of the triad and the primary row is not clear, but the posterior teeth from the triad seem to be homologous, by development and position, to the medial two teeth of the species having one or two regular rows of premaxillary teeth. This character could be considered as part of a single series of transformation with the character 123, given that all the species having a triad, as described by Vari & Harold (2001) also have three recognizable rows of premaxillary teeth, but not all the species with three rows of premaxillary teeth have a recognizable triad. The latter condition was observed by Vari & Harold (2001) and herein confirmed in Creagrutus gephyrus, along with the examined species of Brycon Müller & Troschel, Chalceus Cuvier, and Triportheus Cope. In some of these species there is a tooth situated just anterior to the two medial teeth from the posterior premaxillary row, but this tooth is more slender and similar to the remaining teeth with intraosseous development. The state 1 was coded for Carlastyanax, Piabina, and all the examined species of Creagrutus except C. cracentis and C. gephyrus, which were coded as state 0.
As described by Vari & Harold (2001), in several species of Creagrutus there is a gap between the first and second teeth of the primary premaxillary row that in some species is filled by the anterior tooth of the triad (state 0; Vari & Harold, 2001: figs.  1 and 3; Fig. 3). This condition is absent in other species of the genus (state 1; Vari & Harold, 2001: fig. 2). State 0 could be interpreted as two rows of premaxillary teeth, with the anterior one in zigzag. This character is only coded for the species with a recognizable triad of premaxillary teeth. Among the examined species, only Creagrutus anary Fowler, C. atrisignum, C. meridionalis, and C. taphorni Vari & Harold have state 1.
According to the description presented by Vari & Harold (2001), in Creagutus there is a primary row of premaxillary teeth and a single lateral tooth that is usually situated anterolaterally to the primary row (state 0; Vari & Harold, 2001: figs. 1-3). Medial teeth of the primary premaxillary row of Creagrutus have intraosseous development, while in the lateral ones the development is extraosseous. Thus, the primary row, according to the definition by Vari & Harold (2001) has a double origin. The anterolateral tooth of Creagrutus, treated in this character, has intraosseous development and is considered herein as homologous to some tooth of the anterior premaxillary row of other characids. The species in which all the teeth of the anterior premaxillary row are aligned are coded with the state 1. The state 0 is coded, among examined species, in Bryconamericus exodon, B. emperador, Piabina argentea Reinhardt, and all the species of Creagrutus except C. cracentis. Vari & Harold (2001) coded the state 1 also in C. maxillaris (Myers).
Usually among characids, the anterior process of the maxilla is much more slender than the laminar region of the bone, which is variably expanded (state 0). Vari & Harold (2001) described a thickened and robust posterior region of the maxilla in all species of Creagrutus except C. cracentis and C. maxillaris (state 1; Vari & Harold, 2001: fig. 5B). In the phylogeny of Vari & Harold (2001) this character was proposed as a synapomorphy of Creagrutus, with a reversion in a clade composed of these two species. This character is herein coded as polymorphic in Micralestes stormsi Boulenger, which also has a somewhat thickened anterior region of the maxilla, although not to the same degree as in species of Creagrutus. 372. Flexion on maxilla posterior to site of attachment with premaxilla: (0) absent or not pronounced, maxilla rather straight; (1) flexion pronounced (VH6).
As noted by Vari & Harold (2001), the ascending process of the maxilla is rather aligned with its lamellar region in most characids (state 0). They noted a different state in species of Creagrutus, in which the maxilla has a distinct flexion just posterior to the site of attachment of the ligament connecting this bone with the distal tip of the premaxillary alveolar arm. This flexion gives to the maxilla an angled or curved aspect in lateral view (state 1; Figs. 4-5). Discrete states are difficult to define for this character and the species having morphologically intermediate states were coded as polymorphic. The state 1 of this character refers only to a flexion on the sagittal plane of the maxilla, excluding cases in which the ascending process of the maxilla is curved medially, which are coded with state 0. State 1 is herein coded only for Carlastyanax and the examined species of Creagrutus.

Form and insertion of primordial ligament:
(0) flat and inserted on ascending process of maxilla; (1) round in cross section and inserted on distal half of maxillary lamella (VH8).
The primordial ligament usually originates at the lateral surface of the anguloarticular and the primordial membrane. In most characids, this ligament is flat and inserts on the proximal portion of the ascending maxillary process (state 0). As noted by Vari & Harold (2001), the primordial ligament is stronger, round in cross section, and inserts on middle length of the maxillary laminar portion in the species of Creagrutus (state 1). This state was also herein observed in Carlastyanax.

Form and insertion of primordial ligament:
(0) inserted on middle length of maxillary lamella or dorsal to it; (1) inserted near distal tip of maxilla. (VH9).
In some species of Creagrutus, as observed by Vari & Harold (2001), the primordial ligament is particularly strong and inserts on the region near the distal tip of the maxilla (state 1). This character and the preceding one form a series of nested homologies (sensu Nixon & Carpenter, 2012), and the two characters might alternatively be interpreted as a single additive character with three states. The binary coding of this additive series of transformation follows Mirande (2010), which used this method to facilitate further modifications in the list of characters, considering the great diversity of still non-analyzed species of characids and the intended "modular" nature of that matrix. State 1 was not observed among species examined, and only Creagrutus amoenus Fowler and C. kunturus Vari, Harold & Ortega are coded with that state following observations by Vari & Harold (2001). Thus, this character is informative only in the analyses based on the Extended Matrix.
The primordial ligament originates on the primordial membrane and the anguloarticular in most characids (state 0). Vari & Harold (2001) noted that in most species of Creagrutus the posterior region of the primordial ligament bifurcates posteriorly, attaching to the anguloarticular and also laterally to the quadrate (state 1; Vari & Harold, 2001: fig. 7). In most alestids, as described and illustrated by Zanata & Vari (2005: fig. 21), the primordial ligament is also split ventrally and its ventral division attaches to the quadrate. We examined the alestid Micralestes stormsi, and in this species the primordial ligament attaches only to the anguloarticular. Therefore, this character is coded as 0 in that species and as uncertain in the remaining alestids. Among examined species, state 1 is coded in Creagrutus anary, C. atrisignum, C. cracentis, C. gephyrus, C. meridionalis, and C. taphorni. 376. Ligament between premaxilla and maxilla: (0) weak, attached to a moderately bifurcated premaxilla; (1) strong, attached to a greatly bifurcated premaxilla (VH10).
In the examined characids, there is a ligament between the posterior tip of the alveolar arm of the premaxilla and the proximal region of the ascending process of the maxilla. In most species, that ligament is rather weak and the region of the premaxilla where it originates is moderately notched (state 0). However, as noted by Vari & Harold (2001), this ligament is comparatively strong and round in cross section in Creagrutus and Piabina. Also, the area of origin of this ligament is expanded in species of those genera, with the distal bifurcation of the premaxilla comparatively more pronounced (state 1). The state 1 is also present in Carlastyanax.

Ligament between premaxilla and distal tip of ascending maxillary process:
In most characids, the distal tip of the ascending maxillary process is attached through ligaments only to the lateral wing of the mesethmoid and the posteroventral tip of the premaxilla (state 0). As noted by Vari & Harold (2001: fig. 8B), in Creagrutus there are two additional ligaments that originate at the dorsal margin of the alveolar premaxillary region and insert on the ascending maxillary process. One of those ligaments inserts on the tip of the ascending maxillary process (state 1) and the other inserts on its middle portion (see comments on next character). The ligament that inserts on the anterodosal tip of the maxilla was herein observed only in species of Creagrutus. 378. Ligament between premaxilla and middle portion of ascending maxillary process: (0) absent; (1) present (VH12).
As described in the preceding character, the ascending maxillary process is not connected through ligaments to the alveolar arm of the premaxilla in most characids (state 0), while in the species of Creagrutus there are two ligaments from the premaxilla that inserts on the ascending maxillary process. One of these ligaments inserts on the middle portion of the ascending maxillary process (state 1; Vari & Harold, 2001: fig. 8B; Fig. 6). This ligament is present in Carlastyanax aurocaudatus and the examined species of Creagrutus. 379. Number of teeth on anterior row of dentary: (0) eight or fewer; (1) ten or more (VH17).
The total number of teeth on the anterior row of the dentary is highly variable within the Characidae, but it is usually higher than 10 (state 1). Vari & Harold (2001) noted that in many species of Creagrutus this number is smaller and there are usually fewer than nine dentary teeth (state 0). State 0 was coded among the Stevardiinae in Bryconadenos tanaothoros, Microgenys minuta, Piabina argentea, and all examined species of Creagrutus except C. cracentis and C. gephyrus. Vari & Harold (2001) also observed state 1 in Creagrutus maxillaris. 380. Number of cusps of anterior dentary teeth: (0) three or fewer; (1) five or more (VH18).
The number of cusps of dentary teeth is highly variable among the Characidae, but in most species five or more cusps are present (state 1). As noted by Vari & Harold (2001), the dentary teeth of most species of Creagrutus have only three cusps (state 0). In the Stevardiinae, state 0 was also observed in Microgenys minuta and Knodus pectinatus. Conical teeth were interpreted as having a single cusp, and species with that condition were consequently coded with state 0. 381. Expansion of dentary lateral to anguloarticular: (0) closer to Meckelian cartilage than to articular socket; (1) closer to articular socket than to Meckelian cartilage (VH19).
In most characiforms, the anguloarticular has two processes that are directed anteriorly and anterodorsally. Those processes articulate with the posterior portion of the dentary. The dentary, although firmly overlapping laterally the horizontal process of the anguloarticular, is typically distant from the anguloarticular socket that articulates with the quadrate. In that condition, the posterior margin of the dentary surpasses only slightly the posterior margin of the Meckelian cartilage (state 0; Figs. 3 and 5). In Creagrutus, on the other hand, the dentary broadly overlaps the anguloarticular, and its posterior margin almost reaches the anguloarticular socket (state 1; Vari & Harold, 2001: fig. 9). This character and the 108 of Mirande (2010) describe a series of nested homologies (sensu Nixon & Carpenter, 2012) and both characters represent the binary coding of a single additive transformation series of three states. State 1 was observed only in Creagrutus, examined species of Serrasalmidae, and African species of the Alestidae.
In most characids, there is an anterodorsal process in the anguloarticular overlapping laterally the posterodorsal margin of the dentary. As noted by Vari & Harold (2001), this process is almost vertical (state 0) in most species of Creagrutus, rather than anterodorsally angled (state 1) as in the usual condition of the Characidae. State 0 was also observed in The quadrate typically has a notch, most evident in lateral view, in which the symplectic inserts (state 0). Vari & Harold (2001: fig. 11) described the presence of a dorsally oriented ridge on the quadrate that laterally covers the symplectic (state 1). State 1 was hypothesized as a synapomorphy of Creagrutus by Vari & Harold (2001), with a single reversion in C. lepidus Vari, Harold, Lasso & Machado-Allison. 385. Form of fourth infraorbital: (0) quadrangular, reaching posterior margin of third and fifth infraorbitals; (1) triangular, excluded from posterior margin of infraorbital series (VH31).
The fourth infraorbital is usually quadrangular among examined characids, with its posterior margin as deep as the anterior one (state 0). Vari & Harold (2001) noted that in Piabina argentea the fourth infraorbital is approximately triangular, with its posterior margin angled and not reaching the posterior margin of the third and fifth infraorbitals, consequently being excluded from the posterior margin of the infraorbital series (state 1; Vari & Harold, 2001: fig. 12B). State 1 was also observed in Aphyocharax anisitsi Eigenmann & Kennedy, A. dentatus Eigenmann & Kennedy, and Hasemania nana (Lütken). This character is coded as inapplicable in Carlastyanax aurocaudatus and the other species lacking the fourth infraorbital (Fig. 5).

Parietal branch of supraorbital laterosensory canal:
(0) extended posteriorly to middle region of parietal; (1) absent or reaching only to anterior half of parietal (VH39). In most characids, the parietal branch of the supraorbital laterosensory canal (sensu Di Dario, 2004) runs longitudinally by the posterodorsal region of the frontal and reaches to the posterior margin of the parietal (state 0). This branch is absent or reduced in some species, not reaching the parietal or reaching only the anterior half of this bone (state 1). This character is variable in Bryconadenos tanaothoros, Bryconamericus mennii Miquelarena, Protogino, Filiberto & López, B. rubropictus (Berg), and Cyanocharax obi, which were coded as polymorphic. Carlastyanax aurocaudatus, Creagrutus maracaiboensis, and several non-stevardiin species have state 1 (Fig. 5).
In most characids, the region anterior to the scapular foramen is cartilaginous. That cartilage is limited dorsally by the cleithrum, ventrally by the coracoid, and posteriorly by an anterior projection of the mesocoracoid (state 0). Vari & Harold (2001) noted that in species of Creagrutus examined by them that region is almost completely ossified, with the consequence that this cartilage is small (state 1). That condition was also observed in several other characids, including all the examined species of Creagrutus, but not in Carlastyanax, in which this cartilage is well developed.
The number of scales in the lateral series is frequently used to distinguish between species of several genera of the Characidae. The number of scales is usually higher than 35 in the family (state 0), with several evidently parallel increases of this number in various clades. Vari & Harold (2001) noted that within Creagrutus there is a reduced number of scales in the lateral series of the trans-Andean species C. maracaiboensis and C. nigrostigmatus Dahl, which have less than 33 scales (state 1). This reduced number of scales in the lateral series was also observed herein in Aphyocharax nattereri (Steindachner), Hyphessobrycon elachys Weitzman, Moenkhausia sanctaefilomenae (Steindachner), Pristella maxillaris (Ulrey), Thayeria boehlkei Weitzman, and T. obliqua Eigenmann. A relatively low number of lateral scales in Creagrutus maracaiboensis and C. nigrostigmatus is unique among the analyzed Stevardiinae. (1) four or five median scales (VH52).
Most characids have one or two median scales between the anus and the anal fin (state 0). Vari & Harold (2001) noted that Creagrutus cracentis, C. maxillaris, and C. ungulus Vari & Harold have four or five scales in that region (state 1).

Midlateral body pigmentation:
(0) not pigmented or with a continuous lateral stripe; (1) with a series of dark spots that coalesce during development (VH54).
Most characids lack a broad lateral stripe, or if present, it is regular and homogeneously intense along the lateral of the body (state 0). That stripe is usually silver in live specimens and it turns either diffuse or dark in fixed material, depending on the species and conditions of fixation. State 0 might actually not represent a single condition, but more studies about the histological nature of this pigmentation and the influence of time and process of fixation would be necessary in order to convincingly divide it into more states. Regardless of that question, Vari & Harold (2001) noted that juvenile specimens of Creagrutus amoenus and C. kunturus have dark spots on body sides that coalesce during development, forming a lateral stripe in adults (state 1). This pattern of development of the lateral stripe was not observed in other species of the Characidae. This character is not variable and therefore it is uninformative for parsimony analyses in the Examined Matrix, but it is informative in the Extended Matrix. 391. Total number of vertebrae: (0) 33 or fewer; (1) 34 or more (VH56).
The number of vertebrae is highly variable within the Characidae, but there are usually between 35 and 40 vertebrae (state 1). Vari & Harold (2001) noted that in C. maracaiboensis and C. nigrostigmatus, which are trans-Andean species of the genus, a lower number of vertebrae (less than 34) is present (state 0). This condition was also observed in several other species of the family, but interestingly not in the examined members of the Stevardiinae.

Notochondral mineralizations: (0) absent; (1) present (VH57).
The notochord persists in the region posterior to the uroneurals, between the base of the first and second caudalfin rays, and it is margined by the opisthural cartilage (McDowall, 1999). The notochord is not stained with alizarin in most species, evidencing a lack of mineralization on it (state 0). Vari & Harold (2001) described a condition in which the notochord is mineralized, staining deep red with alizarin (state 1). This state is present in Creagrutus atratus Vari & Harold, C. bolivari Schultz, and C. magoi Vari & Harold by Vari & Harold (2001) and it was not observed herein in other species. In consequence, this character is not variable for the Examined Matrix, but it is informative in the Extended Matrix.
Phylogenetic results. Almost identical phylogenetic results of the Examined Matrix were obtained in a rather wide range of values of K. This range extends from the 7 th to the 10 th value of K (9.37 to 12.31), with two slightly different configurations of 2581 steps each. The first of these trees (obtained in values of K from 9.37 to 11.21) is the most stable according to the measures of sprdiff and tcomp (see Material and Methods). As the differences of the two trees are minimal, both in terms of topology and in measures of stability, the strict consensus of the most parsimonious trees obtained under all these four values of K is proposed as our final hypothesis (Fig. 7).
The relationships obtained after the analysis of the Extended Matrix are comparatively less stable, but much of their instability is found to be within Creagrutus. Most of the analyzed characters are currently unknown for many species of Creagrutus examined by Vari & Harold (2001) and were coded as question marks in the Extended Matrix. This situation produces a high number of most parsimonious trees and a comparatively lower resolution due to the presence of many zero length branches that collapse in the consensus. The low resolution in terms of relationships within Creagrutus, in turn, affected the calculations of tcomp. However, the sprdiff measures show rather stable results in the 9 th to 11 th values of K (11.47 to 13.88). The topology of the most parsimonious trees obtained with the Extended Matrix is almost identical to the ones obtained after the analysis of the Examined Matrix, apart from the inclusion of additional terminals in the latter (Fig. 8).
The extended analysis was performed mainly to evaluate, with all the available previously published evidence, if Carlastyanax is the sister group of Creagrutus, or if it might be included in that genus. The internal relationships of Creagrutus are very complex, as noted by Vari & Harold (2001), and its resolution is beyond the scope of this paper. However, according to the results of our analyses, Carlastyanax is the sister group of Creagrutus in a clade of the Stevardiinae that also includes Bryconamericus exodon, Microgenys Eigenmann, and Piabina (Fig. 7). The synapomorphies of the most inclusive clades of Stevardiinae and the autapomorphies of Carlastyanax aurocaudatus are listed below, along with a more in depth discussion focused on the clades which are more closely related to Carlastyanax. Convergences and reversions are listed only within the Stevardiinae. In the following section, clades are discussed following the nodes in the cladogram depicted in Fig. 7. The number of the characters supporting the monophyly of each node is given in parentheses after its definition, according to the list of characters provided in Supplementary File 2. Values in parentheses after each node indicate (stability/support) indexes, which are expressed as GC values and explained in the Material and Methods section. The GC values are negative in some clades, meaning that there is an alternative clade having a higher frequency in the source trees. This is comparatively more frequent in values of support than in those of stability and are denoted with a "<" before the corresponding number. Support was calculated with K=10.24, corresponding to the 8 th value of K herein explored.
The composition and synapomorphies of the Stevardiinae are the same as those proposed by Mirande (2010), but our results indicate that the genus Carlastyanax should be revalidated. This clade is similar to "clade A" of Malabarba & Weitzman (2003), and is also similar to a clade based on molecular data proposed by Javonillo et al. (2010), which was also named "clade A" in that study. The Stevardiinae, according to the molecular phylogeny presented by Oliveira et al. (2011) includes Bryconamericus emperador and Markiana nigripinnis as the sister group of a monophyletic unit compatible with our Stevardiinae. Although the latter species are not closely related with the Stevardiinae in the analyses presented herein, the congruence in the monophyly of the Stevardiinae among hypotheses generated from morphological and molecular analyses is remarkable given the general low stability and support of most of the internal clades of the Characidae when different kinds of data are analyzed. The close relationship of Markiana with the Stevardiinae was also hypothesized by Baicere-Silva et al. (2011) based on spermatic evidence. This hypothesis is not compatible with our results, in which this genus is related with Astyanax, but the congruence between molecular and spermatic data in this issue is remarkable. Markiana is a morphologically divergent genus that was found to have an unstable position across different analyses by Mirande (2010), and its inclusion in the Stevardiinae should be better evaluated in phylogenetic analyses combining different sources of information.
The position of Hemibrycon at the base of the Stevardiinae was proposed by Mirande (2010). The internal relationships of this clade are different from those proposed in the literature. Two of three synapomorphies previously proposed to this clade were corroborated as such according to our results. In the phylogeny proposed by Oliveira et al. (2011), Hemibrycon taeniurus (Gill), the single species of the genus included in their analyses, is part of a less inclusive clade within the Stevardiinae. This incongruence could be, at least partially, related to the fact that different species of Hemibrycon were analyzed in both studies. In the phylogeny proposed by Mirande (2010), Bryconamericus mennii is included in a clade with Acrobrycon Eigenmann & Pearson, Attonitus Vari & Ortega, Aulixidens, Diapoma Cope, Mimagoniates Regan, and Pseudocorynopoma Perugia. In our analyses, B. mennii is the sister group of this large clade of stevardiins. Although the character 359 (form of the sperm nuclei) is optimized as a synapomorphy of this clade, it should be noted that this character is unknown and hence coded as missing in many of the species within this clade. Thus, both the relationships herein proposed for this clade and the synapomorphic condition for that character in this node should be considered as provisional. The relatively basal position of Cyanocharax alburnus in the Stevardiinae was not explicitly proposed in previous studies. Cyanocharax alburnus lacks most features that were considered by Malabarba & Weitzman (2003) as presumably apomorphic for the genus, with the exception of the presence of only six branched pelvic-fin rays. Cyanocharax Malabarba & Weitzman is polyphyletic according to our results, with C. obi as the sister group of a clade composed of Acrobrycon, Argopleura, Diapoma, Mimagoniates, and Pseudocorynopoma. However, those relationships have a low support, and that is also the case of most basal clades of the Stevardiinae. In the phylogeny proposed by Oliveira et al. (2011), which was based on molecular data, Cyanocharax alburnus is deeply nested within the Stevardiinae, as the sister group of a clade composed of Bryconamericus exodon, Hypobrycon maromba Malabarba & Malabarba, and Odontostoechus lethostigmus Gomes. The phylogeny proposed by Javonillo et al. (2010) included four species of Cyanocharax. In the hypotheses presented by those authors, Cyanocharax is paraphyletic, since it includes Diapoma. A phylogenetic hypothesis for Cyanocharax was published in the description of Cyanocharax obi (Casciotta et al., 2012). In that analysis, the outgroup was composed of unspecified species of Bryconamericus Eigenmann, Diapoma, Knodus Eigenmann, and Odontostoechus Gomes. Casciotta et al. (2012) conducted two different analyses: a morphological one, in which no species of other genera were included to test the monophyly of Cyanocharax, and a Bayesian analysis including only molecular data. In their molecular phylogeny, Casciotta et al. (2012) also obtained Cyanocharax as paraphyletic in terms of Diapoma.
Most species included in this clade were hypothesized as related to Creagrutus and Piabina by Mirande (2010). In the present hypothesis this clade also includes Attonitus, which was proposed as related to Aulixidens by Mirande (2010), and excludes Creagrutus and Piabina. This clade does not include B. exodon, the type species of the genus. Consequently, the species of Bryconamericus included in this clade should be assigned either to Odontostoechus or to a new genus, but more thorough studies including a higher number of species possibly related to this clade are needed.
A close relationship between Bryconamericus agna and the group of species of Knodus in this clade, which includes the type species of Knodus (K. meridae), is proposed for the first time. According to our results, at least three species currently assigned to Knodus, K. breviceps, K. gamma, and K. meridae, form a monophyletic group that would correspond to the "true" Knodus. Node 260 includes three species currently assigned to Knodus, but those species should be transferred to another genus according to our results. However, only a single homoplastic character state, related to maxillary dentition, supports clade 272. Additional characters and examination of more species of Knodus are needed to further support those results.
Three molecular phylogenetic studies included species of Knodus. Calcagnotto et al. (2005) analyzed only one unidentified species of the genus, which according to their results is the sister group of an unidentified species of Bryconamericus. Javonillo et al. (2010) analyzed four presumably new species of Knodus, which form a clade with Bryconadenos and Piabina. That clade is hypothesized to be the sister group of Bryconamericus diaphanus (Cope) (not analyzed herein). Oliveira et al. (2011) included only K. meridae in their analyses, which was hypothesized to be the sister group of Bryconadenos tanaothoros. It is noticeable that two of those studies found B. tanaothoros to be closely related to some species of Knodus. In our analyses B. tanaothoros is the sister group of a clade composed of Aulixidens eugeniae, Knodus heteresthes, and K. pectinatus (see node 260).  Weitzman & Menezes, 1998;Weitzman, 2003) mainly by sharing the presence of a caudal-fin gland. Subsequently, Menezes & Weitzman (2009) observed that the caudal-fin glands of Glandulocauda, Lophiobrycon, and Mimagoniates are morphologically and histologically distinct from those of the remaining genera. They considered those two types of caudal-fin glands as not homologous, divided the Glandulocaudinae, maintaining Glandulocauda Eigenmann, Lophiobrycon Castro, Ribeiro, Benine & Melo, and Mimagoniates in this subfamily, and resurrecting the Stevardiinae to include the remaining genera bearing a caudalfin gland. Mirande (2010) proposed that the genera of Glandulocaudinae and Stevardiinae (sensu Menezes & Weitzman, 2009) form a monophyletic group, deeply nested within a larger clade, which was redefined by him as a more encompassing Stevardiinae.

Synapomorphy
The Glandulocaudinae (sensu Weitzman & Menezes, 1998) was proposed as related to Attonitus, Aulixidens, and Bryconamericus mennii by Mirande (2010), but that hypothesis was not corroborated by the present analysis. In our results, B. mennii is the sister group of a large clade including most analyzed species of the Stevardiinae, Aulixidens is related to some species of Knodus, and Attonitus is included in the node 257 with Nantis, Odontostoechus, and some species of Bryconamericus. One synapomorphy of this clade is the presence of insemination, whose presence is actually unknown in many stevardiins.
This clade includes a group of species currently assigned to Knodus that are not closely related to K. meridae, its type species, but instead form a clade with Aulixidens eugeniae and Bryconadenos tanaothoros. The sister group of this clade (Node 232) includes Cyanocharax obi, Piabarchus analis, and the Glandulocaudinae of Weitzman & Menezes (1998). As noted above, one of the synapomorphies of the Node 233 (including this clade and the Node 232) is the presence of insemination, and probably all the species currently assigned to Knodus with insemination, as K. pectinatus, should be included in the Node 260. However, insemination is unknown in Aulixidens eugeniae and K. heteresthes. Knodus pectinatus was originally described as a species of Bryconamericus by Vari & Siebert (1990) and later on it was tentatively transferred to Knodus  due to the presence of scales covering the anterior region of the caudal fin, a character that traditionally defined that genus. In the present analysis Knodus meridae (the type species of the genus) is included in the node 272 (see above) together with Bryconamericus agna, Knodus breviceps, and K. gamma. Also, Bryconamericus exodon (the type species of Bryconamericus) is included in a separate clade (see node 353). Thus, the inclusion of Knodus pectinatus in Bryconamericus or in Knodus is not supported according to our results. The four species of this clade should be transferred to Aulixidens, which is older than Bryconadenos. Interestingly, in these species, as in Knodus, scales cover the base of the caudal fin. However, this clade has low support and stability, and there are many presumably related species, as the inseminating species of Knodus  that could be analyzed.
As mentioned before, the Glandulocaudinae sensu Weitzman & Menezes (1998) are included in a clade with Cyanocharax obi and Piabarchus analis according to our results. The presence of insemination in the latter two species is unknown. The sister group of the Glandulocaudinae of Weitzman & Menezes (1998) is different in relation to previous hypotheses. In the molecular study of Calcagnotto et al. (2005) a clade formed by Gephyrocharax and Mimagoniates was proposed as the sister group of the remaining Stevardiinae (as treated herein). Javonillo et al. (2010) obtained Mimagoniates as the sister group of the remaining Stevardiinae (as treated herein), among which the Glandulocaudinae (sensu Weitzman & Menezes, 1998) are not monophyletic, with a clade composed of Corynopoma and Gephyrocharax as the sister group of the remaining stevardiins (excluding Mimagoniates). According to Javonillo et al. (2010), Diapoma is included in Cyanocharax, whereas Pseudocorynopoma is the sister group of Bryconamericus stramineus Eigenmann (in their analysis by parsimony), making the Glandulocaudinae (sensu Weitzman & Menezes, 1998) polyphyletic. The Glandulocaudinae of Weitzman & Menezes (1998) is also not monophyletic according to Oliveira et al. (2011), in which Tyttocharax and Xenurobrycon form the sister clade of a large group of species composed of almost all the remaining stevardiins (as treated herein). In the phylogeny presented by Mirande (2010), the Glandulocaudinae of Weitzman & Menezes (1998) is the sister group of Bryconamericus mennii, which in the present hypothesis is included in the Node 237 (see above). This node includes Mimagoniates rheocharis, the single analyzed Glandulocaudinae sensu Menezes & Weitzman (2009) as sister group of the Stevardiinae, as defined by them. This contrast with the results by Mirande (2009Mirande ( , 2010 in which Mimagoniates was obtained to be the sister group of Pseudocorynopoma, resulting in a paraphyletic Stevardiinae sensu Menezes & Weitzman (2009). As in the hypotheses by Mirande (2009Mirande ( , 2010, the Glandulocaudinae and Stevardiinae of Menezes & Weitzman (2009) are deeply nested in a more inclusive clade, which was used to support the redefinition of the Stevardiinae by Mirande (2009).
presented by Mirande (2009Mirande ( , 2010. The relationships of Carlastyanax and Microgenys were not previously addressed in phylogenetic studies, and their placement in a clade that also includes Creagrutus and Piabina is one of the main results of this paper. This clade is stable across all the analyses conducted herein, and has a relatively high support. Piabina is not the sister clade of Creagrutus, but is related to species of Bryconadenos, Bryconamericus, Cyanocharax, Hypobrycon Malabarba & Malabarba, Knodus, Odontostoechus, and/or Piabarchus according to molecular studies (Javonillo et al., 2010;Oliveira et al., 2011). Still according to those studies, Creagrutus is more basal, hypothesized to be the sister group of most species of the Stevardiinae. Those results might be related to the low sampling of genes or absence of more representative species, especially of the highly diverse Creagrutus, but could also indicate that some of the characters proposed by Vari & Harold (2001) that support a sister-group relationship between these genera according to morphology are actually convergences. this relationship, Carlastyanax is revalidated and proposed as a member of the Stevardiinae. The sister group relationship between Carlastyanax and Creagrutus is stable in all the explored conditions in different analyses, and is also well supported according to resampling measures. Some synapomorphies relating those genera, especially the presence of a ligament between the ascending process of the maxilla and the dorsal margin of the alveolar premaxillary ramus, are unique to this clade or highly unusual among the Characidae. (1 > 0) rounded, with posterior notch. Reversed in Creagrutus maracaiboensis.

Results of the Extended Matrix
The most stable results in the Extended Matrix, including all the species of Creagrutus coded from literature, were obtained in the 10 th and 11 th values of K (= 12.59-13.88). The consensus of the most parsimonious trees obtained in these concavities is much similar to the one obtained from the Examined Matrix, except by the position of the species of Knodus and the internal relationships of Creagrutus (Fig. 8).
In this hypothesis the species of Knodus form a monophyletic group, with K. gamma as the sister group of the remaining species of the genus. However, this clade has low stability and support, and therefore this hypothesis should be further tested with the addition of a higher number of species and characters. Within Creagrutus, a clade composed of C. maracaiboensis, C. muelleri, C. nigrostigmatus, C. ouranonastes, and C. peruanus is the sister group of another composed of C. affinis Steindachner, C. crenatus Vari & Harold, C. hildebrandi Schultz, C. lepidus, C. manu Vari & Harold, C. melasma Vari, Harold & Taphorn, and C. pearsoni Mahnert & Géry, and a polytomy including the remaining species of the genus, of which some are grouped in small clades (Fig. 8). Those relationships are similar to the ones proposed by Vari & Harold (2001). Additional resolution in this large clade relative to Vari & Harold (2001) is possibly attributed to the use of implied weighting instead of equal weighting. The large polytomy obtained is almost exclusively due to the lack of unambiguous synapomorphies for most of the internal nodes of the genus, rather than to characterstates conflicts. The resolution of the relationships within Creagrutus is beyond the scope of this paper and the intention of this last analysis is only to corroborate the position of Carlastyanax when all the currently available information is analyzed in a single matrix. Eigenmann (1913) placed the species now recognized as Carlastyanax aurocaudatus in Astyanax in spite of the presence of four teeth in the posterior premaxillary row. Eigenmann later considered that this species should be separated from Astyanax and included in a different genus, based on its coloration, the form of the third infraorbital, and the form of the third dentary tooth (Eigenmann, 1927). However, Eigenmann did not mention the possession of just four teeth in the posterior premaxillary row as something unusual for Astyanax or a species thought to be related to that genus. The creation of Carlastyanax by Géry (1972) was not well supported because it was based on the morphological divergence of A. aurocaudatus and not on the supposed relationships of this species. The degree of morphological difference or similarity in itself is not a reason to take nomenclatural decisions. The single important question is if one or more characters indicate (at least tentatively) phylogenetic relationships between taxa. Astyanax aurocaudatus has all the diagnostic characters of Bryconamericus except the reduced third infraorbital (Fig.  5), although this feature is actually present in some species of the genus (e.g., Bryconamericus rubropictus). The hooked third dentary tooth could also have been considered as an autapomorphy of A. aurocaudatus, and as such, irrelevant to the relationships of the species. However, the creation of Carlastyanax to include A. aurocaudatus by Géry (1972), in the absence of a phylogenetic hypothesis  Fig. 7.

Discussion
for the Characidae, provided a generic name that in the present analysis is supported as valid.
Ruiz-C. & Román- Valencia (2006) considered that the characters supporting the creation of Carlastyanax were "incorrect observations", and thus synonymized Carlastyanax with Astyanax. Those authors reasoned that, as some species of Bryconamericus have five teeth on the posterior premaxillary row (they cited Román-Valencia, 2003), that condition could not be diagnostic of Astyanax (Ruiz-C. & Román- Valencia, 2006: 61). We interpret that their reasoning was that if the presence of five teeth on the posterior premaxillary row is unreliable as a "generic" character state, the same occurs with the presence of four teeth and, thus, the presence of only four teeth in Carlastyanax is irrelevant to its generic assignment. That reasoning is obviously flawed.
Ruiz-C. & Román- Valencia (2006: 61) also stated that the form of the third dentary tooth could not be considered as a diagnostic character of Carlastyanax, because both Astyanax brachypterygium Bertaco & Malabarba and A. cremnobates Bertaco & Malabarba have curved dentary teeth (they cited Bertaco & Malabarba, 2001). However, they did not explicitly stated that Carlastyanax and those two species of Astyanax might be related, a situation that would justify both the inclusion of Carlastyanax in Astyanax and the non-validity of the third dentary teeth as a diagnostic feature of Carlastyanax. In addition, a particularly curved third dentary tooth in those species of Astyanax was not mentioned by Bertaco & Malabarba (2001) or shown in their figures. Although we did not analyze specimens of A. brachypterygium, examined specimens of A. cremnobates do not have that feature. Bertaco & Malabarba (2001: 226, 229) mentioned that "all cusp tips" of the dentary teeth of those species of Astyanax are "slightly curved posteriorly towards inside of mouth". This condition is commonly found in the characids and is actually quite distinct from the condition present on the third dentary tooth of C. aurocaudatus. In our analyses, this condition is an autapomorphy of Carlastyanax aurocaudatus, diagnostic for the genus, but irrelevant for the generic assignment of this species.
Ruiz-C. & Román- Valencia (2006) also discussed some similarities of Carlastyanax aurocaudatus with Astyanax mexicanus (De Filippi), the type species of Astyanax, such as the absence of teeth in the palatine, ectopterygoid, and mesopterygoid. However, those authors did not mention that these features are also shared by most characids. They also said that the putative close relationship of Carlastyanax with some species of Hemibrycon, proposed by Géry (1972), is not supported because they have different "coloration models" ("modelos de coloración", in the original) and "ecological aspects". Regardless of the validity of those remarks, most systematists would not find those attributes as relevant in a taxonomic discussion. Also, as mentioned above, differences themselves are not indicators of phylogenetic relationships and should not be taken into consideration for generic assignments.
Several inconsistencies in the osteological description of Carlastyanax aurocaudatus presented by Ruiz-C. & Román-Valencia (2006) were also detected. According to those authors, the supraorbital bone is present in C. aurocaudatus as a long and narrow structure joined laterally to the frontal. However, the supraorbital is absent in the specimens of C. aurocaudatus we examined, a condition previously noted by Bertaco et al. (2007: 255). We agree with Bertaco et al. (2007: 255) that the "supraorbital" of Ruiz-C. & Román- Valencia (2006) in C. aurocaudatus and some species of Hemibrycon is actually the lateralmost portion of the frontal. Mirande (2009Mirande ( , 2010 found that the supraorbital bone is absent in the Stevardiinae and only present in basal members of the clade assigned to Characidae (e.g., Acestrorhynchinae, Agoniatinae, Bryconinae, Cynodontinae, and Salmininae). Ruiz-C. & Román-Valencia (2006) also reported the presence of a rhinosphenoid "in cartilaginous state" in C. aurocaudatus. In the adult specimens examined by us the rhinosphenoid is absent, and only the ethmoid cartilage is present. Ruiz-C. & Román- Valencia (2006: 57 and figs. 11 and 12) also provided wrong descriptions of the position of the posttemporal, supracleithrum, and cleithrum, including their figures. They stated that in C. aurocaudatus the posttemporal reaches ventrally to the cleithrum, while actually it reaches only to the supracleithrum, as is usual in the Characidae. Ruiz-C. & Román- Valencia (2006) described the supracleithrum as a large bone that is joined to the dorsal margin of cleithrum, but they also stated that the mesocoracoid (spelled as "mesocaracoid" in their paper) articulates with the medial surface of the supracleithrum, when in C. aurocaudatus and in all the remaining examined characids the mesocoracoid articulates with the cleithrum.
Although a comprehensive phylogenetic analysis of the Characidae was not available in the time that Ruiz-C. & Román- Valencia (2006) transferred Carlastyanax aurocaudatus to Astyanax, there was a sort of "agreement" about the monophyly of a clade more or less equivalent to the "clade A" proposed by Malabarba & Weitzman (2003). That clade, later redefined as the Stevardiinae by Mirande (2009Mirande ( , 2010, was proposed based on two putative synapomorphies: the presence of eight branched dorsal-fin rays (vs. nine in most other characids) and the presence of four teeth on the posterior premaxillary row (vs. five or more teeth in most other characids with two rows of premaxillary teeth), which are present in Carlastyanax aurocaudatus. The Stevardiinae of Mirande (2009Mirande ( , 2010 does not include Astyanax, which has five premaxillary teeth in the posterior row and nine branched dorsal-fin rays. Thus, even without considering the published phylogenies and concluding that Carlastyanax is not valid for some reason, this genus has a characters combination that suggested its inclusion in the Stevardiinae and this, in turn, precludes its synonymy with Astyanax. The position of Carlastyanax as the sister group of Creagrutus was not previously explicitly proposed. Vari & Siebert (1990) noted that Knodus pectinatus (referred to in their paper as Bryconamericus pectinatus) shares several features with Carlastyanax aurocaudatus, and therefore a possible relationship between those two species was proposed. Vari & Harold (2001) hypothesized that Piabina is the sister group of Creagrutus, but they did not examine Carlastyanax. According to our results, Piabina is actually the sister group of a clade composed of Carlastyanax and Creagrutus.