Re-validation of Otocinclus arnoldi Regan and reappraisal of Otocinclus phylogeny (Siluriformes: Loricariidae)

Otocinclus arnoldi from the La Plata basin is resurrected from the synonymy of O. flexilis described from the rio Jacui drainage, based on three distinguishing features: the possession of five branched pectoral-fin rays, the larger number of enlarged odontodes on the tip of the parieto-supraoccipital posterior process, and having the prootic involved in the contact with the hyomandibular articular condyle. These species are also compared to O. mimulus, a third species described from the Parana River basin, and the three species are rediagnosed. A reassessment of the phylogenetic relationships of all species of Otocinclus shows a well-supported clade composed of (O. xakriaba ((O. mimulus, O. arnoldi) (O. affinis, O. flexilis))) from the eastern-draining river basins of the Brazilian Shield as sister-group to a clade including all remaining Otocinclus species which are distributed on a wide lowland area of the Amazonas, Paraguay, and Orinoco basins. Otocinclus arnoldi da bacia do rio da Prata e revalidada da sinonimia de O. flexilis, descrito da bacia do rio Jacui, baseado em tres caracteristicas distintivas: a presenca de cinco raios ramificados na nadadeira peitoral, o maior numero de odontodeos hipertrofiados na ponta do processo posterior do parieto-supraoccipital, e por ter o prootico envolvido no contato com o condilo articular do hiomandibular. Essas especies sao tambem comparadas com O. mimulus, outra especie descrita da bacia do rio Parana, e as tres especies sao re-diagnosticadas. Uma nova analise filogenetica de todas as especies de Otocinclus revelou um clado bem suportado composto por (O. xakriaba ((O. mimulus, O. arnoldi) (O. affinis, O. flexilis))) dos rios do escudo Brasileiro que drenam para leste, como grupo-irmao de um clado que inclui todas as demais especies de Otocinclus que sao distribuidas em uma grande area baixa das bacias do Amazonas, Paraguai e Orinoco.


Introduction
Otocinclus flexilis Cope, 1894 was described from the rio Jacuí, Rio Grande do Sul State, Brazil, based on a syntype series with two lots and 17 specimens, collected by Herbert H. Smith in 1882 and deposited in the Academy of Natural Sciences of Philadelphia under catalog numbers ANSP 21622-21626 and ANSP 21756-21767. Cope (1894: 97) diagnosed O. flexilis from O. affinis Steindachner, 1877, mentioning the presence of six branched pectoral-fin rays and describing the color pattern as "light yellowish brown, with a row of about six oblong dusky spots along the lateral line, which become obscure anteriorly. A series of corresponding spots along the dorsal region. Dorsal and caudal fins light colored with numerous dusky spots. A black spot at the base of caudal fin in some specimens".
In the same publication, Otocinclus fimbriatus Cope, 1894 was described from the same type locality of O. flexilis, based on differences "… in the tubercular and fringed lip … more brightly colored and with less numerous lateral spots. The ventral fins are relatively longer, and the dorsal fin originates above their base, and not behind it, as is the case in O. flexilis". Schaefer (1997) examined the syntype series of O. fimbriatus (ANSP 21585-21597, 21752-21755, 17 specimens) and concluded that it does not present consistent differences from O. flexilis and, therefore, O. fimbriatus was synonymized with Otocinclus flexilis, as already suggested by Regan (1904) and Aquino (1996), and maintained by Schaefer (2003).
Otocinclus arnoldi Regan, 1909 was described from "La Plata" based on a single aquarium specimen donated by J. P. Arnold. The rather short original description does not state the number of pectoral-fin rays. Aquino (1996), based on a morphometric and meristic comparison, and also on the color pattern, synonymized O. arnoldi with O. flexilis. More recently, Otocinclus mimulus Axenrot & Kullander, 2003 was described from the río Paraná drainage in Paraguay, being diagnosed from O. flexilis by possessing elevated, enlarged odontodes at the posterior parieto-supraoccipital tip and distinct modal number of premaxillary and dentary teeth and caudal vertebrae.
Otocinclus is the only genus among the Hypoptopomatinae that received strong phylogenetic attention by previous authors. Schaefer (1997) produced the first phylogenetic analysis including all species of Otocinclus known to date. After that, Britto & Moreira (2002) described O. tapirape and reassessed the phylogenetic relationships among Otocinclus species. Axenrot & Kullander (2003) described O. mimulus and again reassessed the phylogeny of Otocinclus species, adding one character to the matrix of Schaefer (1997). Reis (2004) and Lehmann (2006) described O. cocama and O. batmani, but did not provide updated phylogenetic analyses of the genus.
In this paper we present the results of a reassessment of the validity of Otocinclus arnoldi, and reassess the phylogenetic relationships among all Otocinclus species.

Material and Methods
The specimens examined belong to the following institutions: Natural History Museum, London (BMNH); Instituto de Ciencias Naturales, Museo de Historia Natural, Universidad Nacional de Colombia, Bogotá (ICNMHN); Museo de Ciencias Naturales, Guanare (MCNG); Museu de Ciências e Tecnologia, Pontifícia Universidade Católica do Rio Grande do Sul, Porto Alegre (MCP); Museo Nacional de Historia Natural del Paraguay, Asunción (MNHNP); Museu de Zoologia da Universidade de São Paulo, São Paulo (MZUSP); Swedish Museum of Natural History, Stockholm (NRM); Universidade Federal do Rio Grande do Sul, Porto Alegre (UFRGS); Universidade Federal do Rio de Janeiro, Rio de Janeiro (UFRJ); and National Museum of Natural History, Smithsonian Institution, Washington (USNM). Measurements were calculated as interlandmark distances based on homologous landmarks acquired with a video digitizer, using the same set of landmarks as Schaefer (1997). An additional landmark was digitized at the end of the hypural plate and standard length was calculated as the interlandmark distance from this point to the snout tip. The software LMDis (by R. E. Reis, 1996) was used to extract interlandmark distances. Counts and anatomical terminology follow Schaefer (1997), Arratia (2003) and Axenrot & Kullander (2003). Specimens studied were cleared and stained (c&s) using the methods of Taylor & van Dyke (1985). Osteological terminology follows Schaefer (1997) and Arratia (2003).
Principal component analysis (PCA) was used to assess morphometric variation among studied Otocinclus populations. A total of 18 morphometric variables (Table 1) was taken from 70 specimens representing comparable size ranges: 30 specimens from the rio Jacuí basin, type-locality of Otocinclus flexilis (29.4-40.3 mm SL), 27 specimens from the rio Uruguai and lower río Paraná drainage, type-locality of O. arnoldi , and 13 paratypes of O. mimulus from the middle río Paraná basin in Paraguay (30.9-36.3 mm SL). The analysis was performed on the covariance matrix of the 18 log 10 -transformed measurements.
For the phylogenetic analysis we used the original data matrix of Schaefer (1997) with the addition of Otocinclus arnoldi, O. tapirape, O. mimulus, O. cocama, O. batmani, and another undescribed species from the rio Madeira basin provisionally called Otocinclus sp. "madeira". We also included one character from Axenrot & Kullander (2003) on a mimetic association of Otocinclus with the callichthyid Corydoras, and six additional new characters. We submitted the data matrix to 10,000 replications of Random Addition Sequence (RAS) followed by TBR branch swapping using the software NONA (by P. Goloboff, 1993) and WinClada (by Nixon, 2002). Trees were rooted on Microlepidogaster perforatus. All multistate characters were set as unordered. Bremer branch support was calculated with NONA.

Results
A direct comparison of proportional measurements between specimens of Otocinclus flexilis from the laguna dos Patos basin and specimens from the rio Paraná and rio Uruguai drainages (Table 1) reveals no differences, as already demonstrated by Schaefer (1997). The principal component analysis also failed to reveal unambiguous differences among the three populations analyzed. The first principal component included a large proportion of the total variance (78.3%) and all variable loadings were negative and varied little in magnitude, indicating that it represents a general size factor. Plots of factor scores of principal components 2 versus 3 and 2 versus 4 both grouped specimens into three broadly overlapping clusters (Fig. 1). PC 2, 3, and 4 included 7.5, 3.3, and 2.4% of the total variance, respectively.
Lateral trunk coloration in these populations is also similar, as the three groups have either a row of 3-6 distinct dark blotches or a distinct dark stripe, extending from the compound pterotic to the caudal-fin base, or a diffuse mixture of those two color patterns (Figs. 2, 3 and 4). On the other hand, however, there is an important difference in the pectoral-fin ray counts among these fishes, as already demonstrated by Schaefer (1997). All examined specimens from the La Plata basin (285 specimens, including the holotype of O. arnoldi and the 13 paratypes of O. mimulus) have five branched rays in the pectoral fin (except for one specimen each in lots MCP 25254, UFRGS 7180, and USNM 176023 which have six branched rays in one side), and all 226 specimens examined from the laguna dos Patos basin (in addition to the entire syntype series of O. flexilis and O. fimbriatus examined by Schaefer, 1997) have six branched rays in the pectoral fin, except for one specimen in UFRGS 4963 and nine juveniles from MCP 15068, with five rays in one or both sides. Schaefer (1997: 107) described a raised tuft of odontodes forming a raised crest on the parieto-supraoccipital tip as a juvenile character of hypoptopomatines, and regarded the presence of enlarged parieto-supraoccipital crest odontodes in mature adult specimens within Otocinclus as a derived condition representing paedomorphosis, and therefore synapomorphic for O. affinis and O. xakriaba. The same tuft of enlarged, raised odontodes is present in O. mimulus, which was coded as having the derived state in the analysis of Axenrot & Kullander (2003).
Contrary to the three species above, however, mature adult specimens of Otocinclus from the Jacuí and La Plata drainages do not have a raised crest of odontodes on the parieto-supraoccipital tip. However, most of the examined specimens present a patch of hypertrophied odontodes on the parieto-supraoccipital tip which are not raised to form a crest, but instead are laid on the bone surface and cannot be visualized laterally. These odontodes are 2-5 times larger than the surrounding odontodes and are present in both juveniles and adults. The population in the rio Jacuí basin has 0-10 (mode = 0, mean = 3.0) enlarged odontodes, while the fishes from the Uruguai and lower Paraná basins possess 2-23 (mode = 7, mean = 8.3) enlarged odontodes ( Table 2).
The incorporation of the prootic in the contact with the hyomandibular articular condyle was described by Schaefer (1997: 104) as an autapomorphic trait of O. xakriaba. This feature, however, is also present in the Otocinclus populations from the La Plata basin, including O. mimulus, but not in O. flexilis from the rio Jacuí basin, in which only the compound pterotic contacts the hyomandibular articular condyle.
Based on the differences in pectoral-fin count, number of hypertrophied odontodes on the parieto-supraoccipital, and the involvement of the prootic in the contact with the hyomandibular articular condyle, O. arnoldi is resurrected from the synonymy of O. flexilis.
In a similar manner, a direct comparison of proportional measurements and meristics between Otocinclus arnoldi and a series of 13 paratypes of O. mimulus (Table 1) shows no differences, as already mentioned by Axenrot & Kullander (2003: 255  Distribution. Otocinclus flexilis is restricted to the laguna dos Patos drainage basin (Fig. 5).
Distribution. Otocinclus arnoldi occurs in the tributaries of the lower río Paraná drainage, lower and middle rio Uruguai, and the río de La Plata (Fig. 5). flexilis) by having a lateral trunk coloration composed of either a row of 3-6 distinct dark blotches or a distinct dark stripe extending from the compound pterotic to the base of the caudal fin, or a diffuse mixture of those two color patterns (Fig. 4).

Distribution.
Otocinclus mimulus is only known from the río Mondai in Paraguay, a left-bank tributary of the lower rio Paraná (Fig. 5).

Phylogenetic relationships
The phylogenetic position of Otocinclus arnoldi, O. batmani, O. cocama and Otocinclus sp. "madeira" were never investigated. In order to test their phylogenetic position we added to the 27-character data matrix of Schaefer (1997) the character proposed by Axenrot & Kullander (2003) on a mimetic association with Corydoras, and six additional characters (Table 3), which are described below.
Character 28: Mimetic association with a species of Corydoras. According to Axenrot & Kullander (2003) "Otocinclus mimulus, O. flexilis, O. affinis, and O. xakriaba are considered to be mimics of particular sympatric Corydoras species (C. diphyes, C. paleatus, C. nattereri, and C. garbei, respectively)". This interpretation of mimetic association is based on the syntopically co-occurrence and on the shared color pattern of the Otocinclus-Corydoras species pair. We add to this list O. arnoldi, which has the same color pattern of O. flexilis and C. paleatus and also shares the syntopic co-occurrence with C. paleatus (nine of the 23 MCP lots of O. arnoldi were collected syntopically with C. paleatus).
Character 29: Position of pleural ribs. In loricariids the pleural ribs posterior to the well-developed rib of the sixth vertebral centrum are thin and delicate, and variably occur in the first centra posterior to the sixth centrum. In basal loricariids, most neoplecostomines, and most hypoptopomatines the first pleural rib posterior to the sixth centrum is associated with the seventh or eighth centrum (state 0  O. huaorani, O. macrospilus, O. vestitus, O. vittatus, O. tapirape, O. cocama, and O. batmani the first delicate rib is associated with the tenth vertebral centrum (state 2). Otocinclus mura and O. hasemani have no pleural ribs, and this character is thus not applicable. The character is highly variable among the genera of the Hypoptopoma group, but the tribe was coded as having state 0.
Character 30: Shape of the ventral process of the complex centrum (VPCC). In basal loricariids and most hypoptopomatines the VPCC has the shape of a rectangular arch attached to the ventral surface of the complex centrum and contacting the swimbladder capsule by means of a thin, dorsolaterally directed splint. Among species of Otocinclus this state is only shared by O. xakriaba (state 0). In all remaining species of Otocinclus the VPCC lost the dorsolaterally splint, being straight, curved or shaped as a golf-stick, but never forming a complete arch (state 1).
Character 31: Shape of the dorsal-fin spinelet. In basal loricariids the first dorsal-fin spine is transformed in a Vshaped spinelet, which acts with the nuchal plate as a locking mechanism for the second dorsal-fin spine (state 0). In the neoplecostomines and hypoptopomatines, however, the spinelet lost its function as a locking mechanism and is either oval or rectangular and plate-like in shape (state 1), a state shared with Microlepidogaster perforatus and Hisonotus notatus. In all species of Otocinclus, in contrast, the dorsalfin spinelet is V-shaped and the dorsal-fin spine locking mechanism is functional, which represents a reversion in Otocinclus. All species on the Hypoptopoma group completely lost the dorsal-fin spinelet (state 2).
Character 32  remaining species of Otocinclus share the presence of two predorsal plates between the parieto-supraoccipital and the nuchal plate (state 1).
Character 34: Teeth on pharyngeal jaws. Hypoptopomatines generally have well developed teeth arranged in multiple series on both the upper pharyngeal tooth plate and the fifth ceratobranchial (state 0), a condition shared by Hisonotus notatus, Microlepidogaster perforatus and the Hypoptopoma group. All species of Otocinclus share a reduction in the pharyngeal teeth, only having one series of tooth on both the upper pharyngeal tooth plate and the fifth ceratobranchial (state 1).
Besides adding the above characters, we provide modified interpretations of some of the characters as originally arnoldi the lateral ethmoid only slightly expanded medially, covering less than 20% of the nasal capsule when viewed through the nares (Fig. 6a, state 0). Contrastingly, in O. bororo, O. caxarari, O. hasemani, O. hoppei, O. huaorani, O. macrospilus, O. mariae, O. tapirape, O. vestitus, O. vittatus, O. cocama, O. batmani, and Otocinclus sp. "madeira" the subnasal lamina of the lateral ethmoid is moderately exposed and expanded medially in a concave shelf, covering 30-70% of the nasal capsule as viewed through the nares (Fig. 6b, state 1). In the Hypoptopoma group and in O. mura the lateral ethmoid is greatly expanded medially, covering 80-100% of the nasal capsule ( Fig. 6c; state 2). Schaefer's (1997) character 21 is the midlateral dark stripe that can be solid and continuous from the compound pterotic to the caudal-fin base (state 0) or broken in a series of three or more large, diffuse blotches of irregular size and shape (state 1). To this character we added a second state where the midlateral dark stripe is not confluent with the spot at caudalfin base, usually being interrupted one or two plates before the spot (state 2). This later condition is shared by Otocinclus mariae, O. hoppei, O. macrospilus, and Otocinclus sp. "madeira". Otocinclus cocama has a very distinct lateral color pattern, but was coded as having state 1.
Some other characters that deserve comments are: Character 15 is the possession of 23 or fewer lateral plates. Despite that O. cocama and the undescribed species from the rio Madeira possess 21-24 lateral plates, we coded these species as presenting state 1, because the count of 24 plates is rare. Character 18 refers to the number of canal-bearing plates in the anterior field of perforated lateral line plates. Because O. cocama has the lateral line continuous, without a mid-body gap, we coded this character as inapplicable. Finally, the character 22 describes the paired W-shaped marks of the caudal-fin pigmentation. As both O. cocama and O. batmani have one single, wide W-shaped mark on the caudal fin, we added the state 2 for this character to code the condition shared by these two species.
The phylogenetic analysis of this expanded data matrix yielded three maximally parsimonious trees with 70 steps (CI = 60 and RI = 79), the strict consensus of which is presented in Fig. 7. mimulus. One symbol may cover more than one lot or locality. Table 3. Data matrix of characters for Otocinclus species and outgroups. Characters 1-27 from Schaefer (1997), character 28 from Axenrot & Kullander (2003).

Taxon
Character States 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 M. perforatus  Aquino (1996) proposed that Otocinclus arnoldi is a junior synonym of Otocinclus flexilis based on overlapping results in morphometric and meristic characters, and the color pattern. However, she did not include specimens of O. flexilis from the rio Jacuí basin in the morphometric analysis, only comparing the holotype of O. arnoldi with 34 specimens (table 1, lots ILPLA 204 and 207) from the La Plata basin. Furthermore, in her table 1, the holotype of O. arnoldi is erroneously reported as having six pectoral-fin branched rays, what is contrary to our own and Schaefer's (1997: 53) count of five branched rays. The results presented by Aquino (1996), therefore, do not provide compelling justification for considering O. arnoldi as junior synonymy of O. flexilis.

Discussion
Following the synonymy of Aquino (1996), Schaefer (1997: 53) reported on the number of pectoral-fin rays present in O. flexilis, indicating that this feature is variable in this species. The reason for such variability is that specimens from both the rio Jacuí and the La Plata basin were considered to be conspecific. As shown above, all 285 specimens examined of O. arnoldi have five branched rays in the pectoral fin, contrary to all other species of Otocinclus (except O. mimulus), which retain the plesiomorphic state of having six branched rays.
In his revision of Otocinclus Schaefer (1997) reported three lots of O. affinis from the La Plata basin in Argentina (USNM 176023 and USNM 177900) and rio Uruguai in southern Brazil (MCP 9388), significantly extending the geographical range of that species from the coastal rivers of Rio de Janeiro and São Paulo to the La Plata basin. We re-examined these three lots and all 60 specimens have five branched rays in the pectoral fins (except one specimen in USNM 176023 which has six rays in one side). The color pattern is rather faded, but the broad longitudinal stripe or series of blotches can be still seen. For this reason, these specimens are re-identified as O. arnoldi and O. affinis is consequently restricted to the coastal rivers of the Brazilian States of São Paulo and Rio de Janeiro.
The recognition of Otocinclus mimulus as a valid species separated from O. arnoldi is based on slight differences. We preferred this course of action based on the absence of dense sampling of Otocinclus in the area between Santa Fé in Argentina and the río Monday and other tributaries of the rio Paraná in Paraguay. Also, the restricted distribution of O. mimulus is not typical of Otocinclus species but that particular region was already detected as the area of endemism of the cichlid Gymnogeophagus setequedas. It is possible that the distribution of O. mimulus is wider than the río Mondaí alone, but like G. setequedas, it can be restricted to the tributaries of the lower rio Paraná in southeastern Paraguay and the  Brazilian state of Paraná, between the río Mondaí and the Brazilian city of Guaíra.
Despite the addition of six morphological characters, the phylogenetic relationships uncovered by the present analysis did not gain much resolution when compared to the previous analyses (Schaefer, 1991;Britto & Moreira, 2002;and Axenrot & Kullander, 2003 The other Otocinclus clade in the present hypothesis is not well resolved, showing a large amount of polytomy and low internal branch support values. The two most basal species in that clade, however, O. tapirape and O. hasemani, are both endemic to the rio Tocantins, a river emptying near the mouth of the rio Amazonas but still running on rocks of the Brazilian Shield. The remaining of the species are distributed on a wide lowland area of the Amazonas, Paraguay, and Orinoco basins, suggesting that the genus Otocinclus began to differentiate on the watersheds draining the Brazilian Shield uplands and invaded lowlands of the continent only once.