Phylogenetic Systematics, Biogeography, and Ecology of the Electric Fish Genus Brachyhypopomus (Ostariophysi: Gymnotiformes)

A species-level phylogenetic reconstruction of the Neotropical bluntnose knifefish genus Brachyhypopomus (Gymnotiformes, Hypopomidae) is presented, based on 60 morphological characters, approximately 1100 base pairs of the mitochondrial cytb gene, and approximately 1000 base pairs of the nuclear rag2 gene. The phylogeny includes 28 species of Brachyhypopomus and nine outgroup species from nine other gymnotiform genera, including seven in the superfamily Rhamphichthyoidea (Hypopomidae and Rhamphichthyidae). Parsimony and Bayesian total evidence phylogenetic analyses confirm the monophyly of the genus, and identify nine robust species groups. Homoplastic osteological characters associated with diminutive body size and occurrence in small stream habitats, including loss of squamation and simplifications of the skeleton, appear to mislead a phylogenetic analysis based on morphological characters alone–resulting in the incorrect placing of Microsternarchus + Racenisia in a position deeply nested within Brachyhypopomus. Consideration of geographical distribution in light of the total evidence phylogeny indicates an origin for Brachyhypopomus in Greater Amazonia (the superbasin comprising the Amazon, Orinoco and major Guiana drainages), with subsequent dispersal and vicariance in peripheral basins, including the La Plata, the São Francisco, and trans-Andean basins of northwest South America and Central America. The ancestral habitat of Brachyhypopomus likely resembled the normoxic, low-conductivity terra firme stream system occupied by many extant species, and the genus has subsequently occupied a wide range of terra firme and floodplain habitats including low- and high-conductivity systems, and normoxic and hypoxic systems. Adaptations for impedance matching to high conductivity, and/or for air breathing in hypoxic systems have attended these habitat transitions. Several species of Brachyhypopomus are eurytopic with respect to habitat occupancy and these generally exhibit wider geographical ranges than stenotopic species.

pomus and nine outgroup species from nine other gymnotiform genera, including seven in the superfamily Rhamphichthyoidea (Hypopomidae and Rhamphichthyidae). Parsimony and Bayesian total evidence phylogenetic analyses confirm the monophyly of the genus, and identify nine robust species groups. Homoplastic osteological characters associated with diminutive body size and occurrence in small stream habitats, including loss of squamation and simplifications of the skeleton, appear to mislead a phylogenetic analysis based on morphological characters alone-resulting in the incorrect placing of Microsternarchus + Racenisia in a position deeply nested within Brachyhypopomus. Consideration of geographical distribution in light of the total evidence phylogeny indicates an origin for Brachyhypopomus in Greater Amazonia (the superbasin comprising the Amazon, Orinoco and major Guiana drainages), with subsequent dispersal and vicariance in peripheral basins, including the La Plata, the São Francisco, and trans-Andean basins of northwest South America and Central America. The ancestral habitat of Brachyhypopomus likely resembled the normoxic, low-conductivity terra firme stream system occupied by many extant species, and the genus has subsequently occupied a wide range of terra firme and floodplain habitats including low-and high-conductivity systems, and normoxic and hypoxic systems. Adaptations for impedance matching to high conductivity, and/or for air breathing in hypoxic systems have attended these habitat transitions. Several species of Brachyhypopomus are eurytopic with respect to habitat occupancy and these generally exhibit wider geographical ranges than stenotopic species.
Brachyhypopomus is currently represented by 13 valid species, which we list with authors in Table 1. A forthcoming publication by Crampton et al. [3] will confirm the validity of all 13 previously described species, and formally describe 15 additional new species. Because the names of the 15 species under description by Crampton et al. [3] are not yet available (sensu International Code of Zoological Nomenclature [ICZN]), we instead refer to them using the cheironyms listed in bold in Table 1. In accordance with Article 8.3 of the ICZN, 4 th edition, all nomenclatural acts in this paper are disclaimed for the purpose of zoological nomenclature.
Phylogenetic studies of species-level interrelationships in the Rhamphichthyoidea have to date considered only a subset of the species diversity in Brachyhypopomus. In an unpublished doctoral dissertation, Sullivan [24]  Carvalho [22], in an unpublished doctoral dissertation, presented a rhamphichthyoid phylogeny based on a combination of morphological and mitochondrial sequence data (16S rRNA, cytb, and cytochrome oxidase I (COI) genes). This study included 14 species of Brachyhypopomus: B. beebei, B. bennetti (listed as B. sp. "ben"), B. bombilla, B. brevirostris, B. bullocki, patterns of geographical distributions and habitat occupancy. Our analyses combine data from morphology and molecular data, and include all 13 previously-described species of Brachyhypopomus as well as the 15 new species under description by Crampton et al. [3] (Table 1). As outgroups we include single representatives of two non-rhamphichthyoid genera, and seven of the ten rhamphichthyoid genera other than Brachyhypopomus (all genera except Akawaio and Procerusternarchus in the Hypopomidae, and Iracema in the Rhamphichthyidae) (listed in Table 2). Although our analyses focused on obtaining a species-level phylogeny for Brachyhypopomus, the inclusion of multiple rhamphichthyoid genera as outgroups allowed us to comment on the monophyly of Brachyhypopomus, and on phylogenetic interrelationships among rhamphichthyoid genera.

Specimens and collections
We sampled muscle tissue for storage in 96-100% ethanol, or in a buffered solution of 20% DMSO and 0.25 M EDTA at pH 8, saturated with NaCl [25]. All specimens were subsequently fixed in 10% formalin, preserved in 70% EtOH, and assigned lot numbers in biodiversity collections. Specimens for which we collected tissue samples for DNA were euthanized in a 600 mgl -1 solution of eugenol (following the 2013 American Veterinary Medical Association Guidelines for the Euthanasia of Animals) until apnea and EOD cessation. Animal care protocols were approved by the Institutional Animal Care and Use Committee of the University of Central Florida (permits 06-33, 09-36W, 11-39W, and 12-31W).
Specimens from which DNA samples were analyzed were deposited along with tissue samples at the biodiversity collections listed in Table 3, and sequences were deposited in GenBank (also listed in Table 3). Field numbers beginning with the letters WC herein refer to specimens with EODs recorded by the Crampton Lab, and are provided to identify specimens from multiindividual lots. Specimens cleared and stained for osteological analyses are listed in S1 and S2 Appendices.
Specimens subjected to morphological and molecular analyses in this study are deposited at the following biodiversity institutions: Academy of Natural Sciences All field collections conducted by WGRC, CDS, JCW and NRL were authorized by the appropriate national and regional collection and export permits, including: Brazil-via

Geographic and ecological distributions
The geographic ranges and habitat occupancy data presented herein for Brachyhypopomus species are based on the descriptions and redescriptions of all 28 species of Brachyhypopomus provided by Crampton et al. [3], which include distribution maps for 11,750 specimens from 2,642 georeferenced museum lots. Geographic and ecological distributions for outgroup taxa are based on original species descriptions in the literature, and on a recent review of gymnotiform ecology and biogeography [2]. To facilitate the presentation and analysis of biogeographical distributions, collection records listed in this paper are categorized into the following five geographical regions and 14 drainage subunits listed below. These drainage units represent basins, sub-basins, or groups of adjacent small basins selected to together summarize broad patterns of distribution in the genus.

Osteological preparation and nomenclature
The osteology of representatives of 28 species of Brachyhypopomus and nine outgroup species was examined from cleared and stained (CS) specimens; see list of specimens in S1 and S2 Appendices. Specimens of reproductive size for each species were examined to avoid the inclusion of juvenile characters, unless stated otherwise. Radiographs served as supplementary sources of data for some species. Specimens were cleared and counterstained for cartilage and bone using the method outlined by Taylor & Van Dyke [26]. In some specimens with weak ossification, bones were stained with alizarin red in ethanol solution instead of KOH solution [27]. The pectoral girdle, suspensorium, and components of the head were removed following Weitzman [28]. Osteological nomenclature and homology follow de Santana & Vari [29] and Hilton et al. [30], except for lateral line system nomenclature, which follows Arratia & Huaquin [31].

Morphology-based phylogenetic reconstruction
We applied parsimony analysis to a matrix of 60 morphological characters (Table 4) from 37 terminal taxa: 28 ingroup species of Brachyhypopomus (Table 1) and nine outgroup species ( Table 2). The outgroup species match those in our molecular phylogenetic reconstruction and represent seven of the ten rhamphichthyoid genera outside Brachyhypopomus (Gymnorhamphichthys, Hypopomus, Hypopygus, Microsternarchus, Racenisia, Rhamphichthys, Steatogenys). Three genera, Iracema, Akawaio, and Procerusternarchus were not included because cleared and stained specimens and molecular data were not available. Characters were chosen primarily to elucidate inter-specific phylogenetic relationships within Brachyhypopomus. However, we included some autapomorphic (non parsimony-informative) characters in Brachyhypopomus, since these may be indicative of synapomorphies with the addition of newly-discovered taxa in the future. All characters were binary, except for character 48, which is multistate and coded as unordered.
We rooted our phylogenetic reconstructions with Gymnotus, based on Albert [20], which places the Gymnotidae as the sister taxon to all remaining gymnotiforms; although see alternative conclusions in Triques [32] and Alves-Gomes et al. [21]. We subjected the character matrix to a heuristic search in PAUP Ã [33] using the default options (stepwise addition, simple additional sequence, branch-swapping TBR, Max trees at 1000, and branches collapsed if maximum length is zero). We generated character diagnoses and synapomorphy lists, and performed tree manipulations using PAUP Ã and Mesquite 3.04 [34]. Ambiguous character distributions were resolved using accelerated transformation (ACCTRAN), which maximizes reversals over parallelism [35].

Sternopygus astrabes
The amplification of rag2 for the majority of taxa was accomplished using the primers RAG2JF1 (5'-TGCTATCTTCCACCACTGCGVTGCC-3') and RAG2JR1 (5'-TCATCYTCCT CATCKTCCTCATTGTA-3') designed for this study. For some taxa, additional new amplification primers were designed and used (sequences available on request from NRL). PCR reaction Outgroups: Phylogenetic Systematics of Brachyhypopomus volumes and concentrations followed those outlined for cytb, except that 2-5μl of DNA was used in each reaction. Thermal cycling conditions for rag2 used a touchdown protocol of one cycle of initial denaturation at 95°C for 30s, followed by denaturation, 58°C, 56°C, 54°C, 52°C, for two cycles each, then 50°C for 32 cycles annealing, followed by extension at 72°C for 90s. PCR products were purified using QIAGEN PCR purification kits (QIAGEN), and sequenced at the Centre for Applied Genomics facility at SickKids Hospital, Toronto. Phylogenetic analysis. We edited and aligned sequences using Geneious Pro v5.5.6 [36]. For both rag2 and cytb, alignment was trivial and no insertions/deletions were detected. DNA sequences were concatenated by individual and combined with the morphological dataset to produce a total evidence matrix of 2131 characters for 105 operational taxonomic units. We analyzed these data using both Bayesian and parsimony approaches. Bayesian analyses allow the use of more complex models of molecular evolution and assessment of node support across a distribution of most-probable trees.
Using MrBayes v3.2.2 [37,38], we conducted a Bayesian Inference (BI) phylogenetic analysis of (1) each gene separately, (2) both genes combined, and (3) both genes combined with morphological data (total evidence). For these analyses, we employed the best-fit substitution models for each partition selected by PartitionFinder v.1.1.1 [39], with the GTR+G model for Rag2, the GTR+I+G model for cytb, and the Mkv model for morphology. For each analysis, we conducted two independent MrBayes runs, each with four chains for 20 million generations, sampling every 1000 generations to ensure standard deviation of split frequencies were below 0.01 and potential scale reduction factors were close to 1.0. For combined analyses, partitions were unlinked. MrBayes runs were inspected in Tracer v1.6 [40] to ensure convergence of parameter estimates, and we confirmed that the Effective Sample Size (ESS) values of all parameters were well above 200. The first 25% of each run was discarded as burn-in, and we combined the remaining trees and parameter estimates to determine posterior distributions. Trees and posterior probabilities for nodes were inspected using FigTree 1.4.0 [41].
We conducted parsimony analysis of the total evidence dataset using PAUP Ã [33]. We defined three data partitions: rag2, cytb, and morphology, and used the parsimony-based incongruence length difference test (ILD or partition homogeneity test of PAUP Ã ) to explore the phylogenetic congruence of these partitions [33,42]. Pairwise ILD comparison between rag2 and cytb partitions indicated congruence (p > 0.05), but pairwise comparisons of morphology with each molecular partition (cytb and rag2) indicated incongruence (p < 0.05). Combining incongruent data partitions can increase the accuracy of phylogenetic reconstruction [43,44]. Therefore, we proceeded with total evidence analysis (all partitions combined), but also assessed the phylogenetic hypotheses based on individual partitions.
For all analyses, we used the heuristic search algorithm with 1000 replicates of random addition of taxa, and TBR branch swapping. All trees were rooted using Gymnotus. Bootstrap values [45] were calculated in PAUP Ã using the heuristic search option (1000 replicates, 10 random taxon additions), and decay indices (Bremer support values) [46]-were calculated using the program TreeRot.v3 [47].
We used maximum likelihood (ML) to reconstruct ancestral character states for three ecological characters and geographic distributions (see below). We reconstructed characters on the BI total evidence tree using Mesquite v. 3.04 [34] using the Mk model of Pagel [48], with branch lengths set to 1.

Illustrations & photography
Camera lucida tracings of micro-dissected osteological structures were made with a drawing tube attached to a Meiji Techno RZ stereomicroscope. These tracings were used in combination with digital photographs of the equivalent structure (taken through the stereoscope) to add stipple-texture in the final osteological illustrations. All digital rendering was conducted in Adobe Illustrator 2014.1.1 (Adobe Corporation, San Jose, CA). Some illustrations (where noted in figure captions) were prepared from the right side of dissected specimens and are inverted left-to-right to present anterior features in the traditional left position. Photographs of cleared and stained specimens were taken with a video camera or Nikon Coolpix P5100 digital camera attached to the Meiji microscope, or on an illuminated light table with a Nikon Coolpix 5100 camera attached to a Nikon EZ-Micro field microscope at x20 magnification.  2, 3 and 4 show BI phylogenies for Brachyhypopomus based on cytb, rag2, and cytb+rag2 combined, respectively. For all trees, terminal two-letter code in labels of ingroup specimens indicate drainage unit (see 'Geographic and ecological distributions' in Materials and Methods). These topologies are highly concordant, showing that both genes provide useful phylogenetic information. As expected, branch lengths are generally longer for the mitochondrial cytb gene compared to the nuclear rag2 gene; mitochondrial protein-coding genes generally evolve more quickly than nuclear protein-coding genes. Cytb typically provides greater resolution of relationships within species and between closely related species.

Total evidence phylogenetic reconstruction
Our analyses of the total evidence matrix comprising the combined data from morphology, cytb, and rag2 yielded the trees shown in Fig 5 (Bayesian inference) and Fig 6 (parsimony). These trees include two to seven individuals for each sequenced species, and one individual each of B. sp. "arrayae" and B. sp. "menezesi"-for which only morphological data are available.
In Fig 7 we provide a simplified cladogram based on the Bayesian total evidence phylogram depicted in Fig 5, and use this as our preferred topology for discussing morphological character state transitions, ecological characters, and phylogenetic patterns of geographical distributions in Brachyhypopomus (see below). We compared optimizations on the total evidence parsimony tree (not shown), and our inferences are robust to different tree-building methods. Terminal branches in Fig 7 represent species (28 Brachyhypopomus and 9 outgroup species) and we use an alphabetical scheme (A-W) to label clades that are well-supported-defined arbitrarily as node support exceeding 0.88 Bayesian Posterior Probability (PP). Clades that are labeled numerically (1-3) are those for which we have less confidence (PP < 0. 88), and which are more likely to change in future analyses. We recognize the limitations of allocating a cut-off between well-supported and poorly-supported nodes, but make the distinction as a heuristic   Table 3 for list of sequenced specimens. Terminal two letter codes refer to the drainage units described in 'Geographic and ecological distributions' (Materials and Methods).   Table 3 for list of sequenced specimens. Terminal two letter codes refer to the drainage units described in 'Geographic and ecological distributions' (Materials and Methods).   Table 3 for list of sequenced specimens. Terminal two letter codes refer to the drainage units described in 'Geographic and ecological distributions' (Materials and Methods). doi:10.1371/journal.pone.0161680.g004 Phylogenetic Systematics of Brachyhypopomus device to simplify our discussions. In Fig 7 we also label some of the well-supported clades as species groups. These are used to facilitate our discussions, not to imply equal taxonomic rank.

Morphological character descriptions and analyses
Here we discuss the 60 morphological characters used for phylogenetic analysis, ordered by discrete body systems in an approximately anterior to posterior sequence. For each character we provide a summary description of the character states and a discussion of the distribution of character states among the ingroup and outgroup taxa. For each character we list the consistency and retention indices for each character based on the topology for the Bayesian total evidence analysis (summarized in Fig 7). A detailed synapomorphy scheme for morphological characters in Brachyhypopomus, also based on Bayesian total evidence topology, is provided in S3 Appendix. The synapomorphy scheme indicates character state changes at each node, and indicates whether they are ambiguous or unambiguous, and homoplastic or non-homoplastic.
An as-yet undescribed bone in the Rhamphichthyoidea, located above the maxilla, was documented by de Santana and Crampton [49] as part of a discussion of the homology of the infraorbital series, antorbital, and associated structures in Hypopygus. This bone has elsewhere been incorrectly identified as the antorbital [18,20,22,50]) or first infraorbital [19,32] In gymnotiforms the antorbital is located ventral to the cavity of the posterior naris and is not to be confused with the undescribed bone that we discuss in Characters 1 and 2. The antorbital is present (state 0, Fig 9A; Albert et al. [53]: 385, 386, figs. 4,6) in all species of Brachyhypopomus except B. bennetti and B. walteri, and is also present (state 0) in Gymnorhamphichthys, Gymnotus, Hypopomus, Microsternarchus, Rhamphichthys, and Sternopygus. The antorbital is absent (state 1; Fig 9B) in B. bennetti and B. walteri, and also Hypopygus, Racenisia, and Steatogenys. The antorbital of Brachyhypopomus (where present) and Gymnorhamphichthys, Gymnotus, Hypopomus, Microsternarchus, Rhamphichthys comprises a vertically oriented ossified or  Table 3 for list of sequenced specimens. Terminal two letter codes refer to the drainage units described in 'Geographic and ecological distributions' (Materials and Methods). The infraorbital canal opening is located at a vertical to and slightly anterior to the sphenotic spine (state 0; Fig 10A) in all species of Brachyhypopomus, and in Gymnorhamphichthys, Gymnotus, Hypopomus, Hypopygus, Rhamphichthys, Steatogenys, and Sternopygus. Conversely, the infraorbital canal opening is distinctly anterior to the sphenotic spine (state 1; Fig 10B) in Microsternarchus and Racenisia. In taxa exhibiting character state 1, the distance between the infraorbital canal opening and sphenotic spine is contained approximately one to one and a half times in the length of the sphenotic spine ( Fig 10B). 6. Occurrence of parietal branch of the supraorbital canal. (0) present; (1) absent (CI = 0.20; RI = 0.33).
The parietal branch of the supraorbital canal, which runs over the frontal towards the parietal, is present (state 0; [49]: 1104, fig 2) in the adults of all Brachyhypopomus species except B. sp. "alberti", B. sp. "arrayae", B. sp. "belindae", B. sp. "hamiltoni", and B. sp. "verdii", and is also present (state 0) in most adult specimens of Gymnorhamphichthys, Gymnotus, Hypopomus, Microsternarchus, Rhamphichthys, Steatogenys, and Sternopygus. The parietal branch of the supraorbital canal is absent (state 1; Triques [32] The parietal branch of the supraorbital canal over the frontal is positioned at a vertical above the infraorbital canal aperture (following the nomenclature of Arratia & Huaquin [31]) (state 0) in all Brachyhypopomus species which possess the parietal branch of the supraorbital canal, and in Gymnorhamphichthys, Gymnotus, Hypopomus, Rhamphichthys, Steatogenys, and Sternopygus. Conversely, the parietal branch of the supraorbital canal is located distinct posterior to the infraorbital canal aperture (state 1) in Microsternarchus. This character is not applicable to B. sp. "alberti", B. sp. "arrayae", B. sp. "belindae", B. sp. "hamiltoni", B. sp. "verdii", and  Table 3 for list of sequenced specimens. Terminal two letter codes refer to the drainage units described in 'Geographic and ecological distributions' (Materials and Methods).  also Hypopygus and Racenisia, which lack the parietal branch of the supraorbital canal in adults (see Character 6).
The parietal branch of the supraorbital canal is included in (fused to) the frontal (state 0) in all species of Brachyhypopomus which possess a parietal branch of the supraorbital canal (except B. sp. "flavipomus") and in Gymnorhamphichthys, Gymnotus, Hypopomus, Microsternarchus, Rhamphichthys, Steatogenys, and Sternopygus. The parietal branch of the supraorbital canal consists of an independent tube over the frontal (state 1) in Brachyhypopomus sp. "flavipomus". This character is not applicable to B. sp. "alberti", B. sp. "arrayae", B. sp. "belindae", B. sp. "hamiltoni", B. sp. "verdii", and also Hypopygus and Racenisia, which lack the parietal branch of the supraorbital canal in adults (see Character 6).
The supraorbital canal (not including its parietal branch) is included in (fused to) the frontal (state 0) in all species of Brachyhypopomus, except B. sp. "flavipomus" and B. sp. "verdii", and is also included in the frontal (state 0) in all outgroups except Steatogenys. The supraorbital canal is independent from the frontal (state 1) in B. sp. "flavipomus" and B. sp. "verdii", and in Steatogenys.
Neurocranium. 10. In the Rhamphichthyoidea, the lateral ethmoid, where present, has two distinct forms. It is narrow and tube-shaped in its mid-section but distinctly wider at its ventral and dorsal portions, resembling a bow-tie, its width reaching approximately 50% of its length (state 0; Mago-Leccia   The homology of the branchiostegal rays in gymnotiforms was discussed by de Santana & Vari [29]. The first, anterior-most, of five branchiostegal rays, which is attached to the anterior portion of the ceratohyal, is present (state 0; Mago-Leccia [19]: The homology of the first basibranchial in gymnotiforms is disputed; see for example Triques [32] and Hilton et al., [30]. The cartilaginous or ossified first basibranchial is located ventral to the basihyal, between the first hypobranchials. We noted that the first basibranchial is present and cartilaginous (state 0; de La Hoz & Chardon [58]: 30-31, figs. 18-19) in all species of Brachyhypopomus except B. brevirostris, and is also present and cartilaginous (state 0) in all outgroups. In contrast, the first basibranchial is ossified (state 1) in B. brevirostris.
Specimens of small juveniles (< ca. 50 mm TL) of Brachyhypopomus were only available for the species listed above, and yet all exhibited teeth on the premaxilla. We suspect that this character may be more widespread in the genus but are as yet unable to confirm this. Therefore, we code other species of Brachyhypopomus as unknown for this character. We reported the absence of teeth on the premaxilla in both adults and small juveniles (<50 mm TL) (state 1) in the following outgroups: Hypopygus, Microsternarchus, Racenisia, and Rhamphichthys (see S1 and S2 appendices for size range of examined cleared and stained specimens). Small juveniles of Gymnorhamphichthys, Hypopomus, and Steatogenys were unavailable for analysis, and are coded unknown. 47. Form of descending process of maxilla. (0) broad in adults only, or broad throughout ontogeny; (1) narrow in all ontogenetic stages (CI = 0.33; RI = 0.78).
The blade on the posterior portion of the descending process of maxilla is broad (state 0; Fig  13A) throughout ontogeny in B. bombilla, B. diazi, B. sp. "menezesi", B. occidentalis, B. sp. "palenque", B. sp. "regani", B. sp. "sullivani", and in Hypopomus, Microsternarchus and Racenisia. We noted that the blade on the posterior portion of the descending process of maxilla is broad in adults of many of these taxa, while narrow in juveniles, suggesting that the blade broadens during ontogeny. The posterior portion of the descending process of the maxilla is narrow at all ontogenetic stages (state 1; Fig 13B)  The preopercular sensory canals, when present, are incised in the preopercle (state 0, Fig  14A)  Mago-Leccia [51] and Albert [20] commented that members of the Rhamphichthyoidea are recognized by the absence of teeth on the oral jaws; although Fernández-Yépez [64] cited the presence of teeth on the dentary of Microsternarchus bilineatus. We observed the presence of As with the presence of premaxillary teeth, we suspect that the presence of teeth on the dentary may be more widespread in post-larval specimens and small juveniles of the genus. Nonetheless, due to the rarity of identifiable small juvenile specimens in collections, we are as yet unable to confirm this. Therefore we coded all species of Brachyhypopomus for which small juveniles are currently unavailable as "unknown". We noted the absence of dentary teeth (state 1) in both adults and small juveniles (< 42 mm) of Hypopygus, Microsternarchus (contrary to Fernández-Yépez, 1968), Racenisia, and Rhamphichthys (see S1 and S2 Appendices for size ranges of examined cleared and stained specimens). Small juveniles were not available for Gymnorhamphichthys, Hypopomus, and Steatogenys and we therefore coded these taxa as "unknown". 51. Form of dorsoposterior portion of dentary. (0) dorsoposterior portion of dentary straight and even (1) dorsoposterior portion of dentary uneven, or with hook-like process (CI = 0.40; RI = 0.57).
The posterior portion of a small independent, undescribed ossification (bone) above the supraoccipital and anterior portion of neural complex is clearly separated from the supraoccipital (state 0; Fig 16A) in all species of Brachyhypopomus except B. brevirostris, B. bullocki, B. sp. "cunia", and B. sp. "hendersoni", and is also separated from the supraoccipital (state 0) in Microsternarchus and Racenisia. Conversely, this small bone overlaps with the supraoccipital (state 1; Fig 16B) in B. brevirostris, B. bullocki, B. sp. "cunia", and B. sp. "hendersoni". This character is not applicable to taxa for which this small independent ossification is absent, i.e.  414, fig 7). The ascending process of the endopterygoid is absent in Gymnorhamphichthys. This ascending process on the endopterygoid does not contact the orbitosphenoid (state 0; Fig 17A) in B. beebei, B. brevirostris, B. sp. "cunia", B. diazi, B. draco, B. sp. "flavipomus", B. gauderio, B. janeiroensis, B. jureiae, B. occidentalis, B. sp. "palenque", B. pinnicaudatus, and B. sp. "verdii", and in Gymnotus, Hypopomus, Hypopygus, Microsternarchus, Racenisia, and Steatogenys, or is absent (in Gymnorhamphichthys) . The ascending process on the endopterygoid forms a contact with the orbitosphenoid (state 1) in Brachyhypopomus sp. brevirostris. We examined both this cleared and stained specimen and ethanol-preserved specimens from the same lot, and identified them as B. beebei. We noted that in some specimens the disk-like ossification of the palatoquadrate cartilage is present, but relatively hard to discern due to poor uptake of stain. This may explain why Sullivan [24] did not observe this ossification in some species of Brachyhypopomus (B. brevirostris, B. bullocki, B. bombilla, B. sp. "regani", B. sp. "sullivani"). Among the Rhamphichthyoidea accessory electric organs (AEOs) are known in Steatogenys species, which possess a paired mental and humeral AEO [66,67], and Hypopygus, which possess a paired post-pectoral AEO [49]. A paired AEO overlying the operculum and lying immediately under the skin was noted by Sullivan [24] and Carvalho [22] in the genus. The paired opercular AEO is absent (state 0) in all Brachyhypopomus species except B. bombilla, B. sp. "menezesi", and B. sp. "regani", and is also absent (state 0) in all outgroups. Conversely, a paired opercular AEO is present (state 1; Sullivan [24]: 322, fig 54; Carvalho [22]: 177, fig 37) in Brachyhypopomus: B. bombilla, B. sp. "menezesi", and B. sp. "regani". In all three of these species, the AEO is an inverted U-shaped structure which originates near the anus and extends to approximately half-way up the head. It widens from a stalk-like ventral portion to a wider distal portion. The opercular AEO in all three species appears to represents a continuation of the electrocytes and associated gel-like matrix of the hypaxial organ, which extends anterior to the anal and urogenital pores and divides near the isthmus into the paired AEO. The AEO is a peduncular structure comprising translucent oblong or polygon-shaped electrocytes resembling those in the main hypaxial organ. The electrocytes are arranged irregularly in approximately three vertically oriented series; each series comprising some 6-10 electrocytes. The entire AEO is overlain by a thin layer of translucent skin. In B. sp. "menezesi" and B. sp. "regani", the skin overlying the AEO possesses minimal chromatophores, such that the AEO and its margins are clearly visible as a pale patch. In contrast, the skin overlying the AEO of B. bombilla often exhibits a higher density of chromatophores, which occlude the outline of the organ. This probably accounts for why Loureiro & Silva [68] failed to mention the AEO in their description of B. bombilla.

Geographic Distributions
The . This approach mirrors that adopted by studies of the biogeography of two other widely distributed gymnotiform taxa: Gymnotus [53,69], and Sternopygus [61]. The first four regions correspond to cis-Andean drainages (those to the east and south of the Andes). Region 5 groups trans-Andean basins (those to the west and north of the Andes).

Ecological Distributions
Brachyhypopomus species are restricted to shallow-water ecosystems. These divide into two categories: floodplain systems subject to a predictable seasonal inundation cycle, and terra firme systems lying above the extent of seasonal flooding. In floodplain habitats, Brachyhypopomus live and breed in the root mats of floating rafts of grasses and other macrophytes. In terra firme systems Brachyhypopomus usually occur in and around aquatic plants, marginal root mats, and submerged leaf litter, debris, or in the case of shield and piedmont streams, rocks and stones. Brachyhypopomus, and other hypopomids, sensu Maldonado-Ocampo [14], are conspicuously absent from the benthos of large, deep, river channels, where Apteronotidae, Sternopygidae, and some Rhamphichthyidae are abundant [70]. Depending primarily upon catchment geology and geomorphology, Neotropical freshwaters exhibit substantial variation in water chemistry, including variation in conductivity and dissolved oxygen-parameters that are known to influence the localized distribution of Brachyhypopomus, and other electric fish [2,5,8,71,72].
As summarized in Crampton [22], Neotropical floodplain systems are typically of Quaternary origin and flank major rivers along their entire lowland courses-forming mosaics of lakes, Phylogenetic Systematics of Brachyhypopomus channels, and seasonally flooded forest or grassland. Floodplains divide broadly into three major categories. 1. Whitewater floodplains: these flank high conductivity (ca 60-300 μScm -1 ) sediment-rich rivers of Andean origin (e.g. the rio Paraguay, Apuré, Marañon, Ucayali, Juruá, and Madeira). 2. Blackwater floodplains: these flank low conductivity (ca 5-30 μScm -1 ), sediment-poor humic-stained blackwater rivers that derive from forested lowland Paleogene-Neogene formations (e.g. the rio Japurá, Tefé, Negro, Uatamã, and Arapiuns). 3. Clearwater floodplains: these flank low conductivity, sediment-poor (ca 5-30 μScm -1 ) clearwater rivers of shield origin (e.g. the rio Tapajós, Xingú, and Tocantins). Whitewater floodplain waters become anoxic or severely hypoxic during the flood period as a consequence of the decomposition of accumulated leaf litter and other organic debris in inundated forests. In contrast, blackwater and clearwater floodplains waters usually remain well oxygenated through the year (> 2.0 mg/l), primarily due to the small extent of their inundated forests in comparison to whitewater floodplains (and consequently lower rates of deoxygenation from decomposing organic debris) [2,72].
The distributions of Brachyhypopomus (and three other hypopomid species) among the floodplain and firme systems described above are summarized in Fig 19A. Nine species of Brachyhypopomus are eurytopic-occupying both floodplain and terra firme systems. Nine are specialized to river floodplain systems. Ten species are endemic to terra firme systems. We explore these habitat distributions in the phylogenetic context in Fig 19A. Here, on the total evidence phylogeny, we optimize exclusive occurrence in floodplains, exclusive occurrence in terra firme streams, and eurytopy as three character states.
The occurrences of Brachyhypopomus species (and three other hypopomid species) in low conductivity systems (ca. 5-30 μScm -1 ) and high conductivity systems (ca. 60-500 μScm -1 ) are presented in Fig 19B. Nine species of Brachyhypopomus are eurytopic with regard to conductivity. Twelve are endemic to low conductivity systems, and seven are endemic to high conductivity systems. In Fig 19B we also explore distributions with regard to conductivity in the phylogenetic context.
In Fig 20 we classify Brachyhypopomus species (and three other hypopomid species) as either "known to occur in habitats subject to intermittent or perennial hypoxia" (< 0.5 mgl -1 ) (sixteen species of Brachyhypopomus) or "restricted to normoxic habitats" (always > 0.5 mgl -1 ) (ten species of Brachyhypopomus). Here we also optimize these character states onto the total evidence phylogeny for hypopomids.

The Monophyly of Brachyhypopomus
Bayesian analysis of the cytb and rag2 genes separately (Figs 2 and 3), both genes combined (Fig 4), and both genes combined with morphological data (total evidence) (Figs 5 and 7) all provide strong support for a monophyletic Brachyhypopomus (nodal support by Bayesian Posterior Probabilities in each case = 1). Parsimony total evidence phylogenetic reconstruction (Fig 6) also provides support for a monophyletic Brachyhypopomus (nodal bootstrap support = 98%, Bremer support = 12).
In contrast, our phylogenetic analysis based on morphological characters alone reconstructed a paraphyletic Brachyhypopomus, with Microsternarchini at a deeply nested position in the genus, as part of a clade also including B. sp. "batesi", B. sp. "benjamini", and B. sp. "provenzanoi" (sp. "batesi" species-group) (Fig 1). We suspect that the placement of the Microsternarchini in our morphology-based phylogeny results from the sharing of homoplastic characters associated with stream-dwelling and diminutive size in the Microsternarchini and sp. "batesi"-group (see 'Phylogenetic interrelationships within Brachyhypopomus', below), and is therefore incorrect. Only one morphological character qualifies as an unambiguous, unreversed morphological synapomorphy for Brachyhypopomus, and is therefore of diagnostic value: the derived presence of a disk-like ossification in the anterior portion of the palatoquadrate cartilage in adult specimens (character 57; Fig 17). Lack of ossification of the palatoquadrate is the ubiquitous condition in all rhamphichthyoid outgroups included in our analyses. Although we did not include the other known rhamphichthyoid genera Iracema, and Procerusternarchus in our dataset, it appears that these taxa also lack ossification of the palatoquadrate. X-ray computed tomography images of the suspensorium of Iracema caiana and Procerusternarchus pixuna show no Fernandes et al. [18] : 101, fig 3). Photographs of cleared and stained specimens of Akawaio penak (ROM 83884) show the palatoquadrate cartilage is also unossified in this genus. Our conclusion regarding the monophyly of Brachyhypopomus is contingent on the taxa we included in this analysis; collection of full morphological and molecular character data for Iracema, Akawaio, and Procerusternarchus will be necessary for a comprehensive test of Brachyhypopomus monophyly.
Based on the Bayesian total evidence topology (Figs 5 and 7), morphological characters 22, 32, 41, 46, and 50 are also optimized as synapomorphies for Brachyhypopomus, but have limited diagnostic value. In some cases these characters are reversed in the ingroup and/or outgroup taxa, in some taxa the characters are inapplicable, and in some cases complete character state data are not yet available for all taxa. We speculate that two of these synapomorphies could represent additional diagnostic characters for Brachyhypopomus: the presence of premaxillary teeth at some stage in ontogeny (Character 46), and the presence of dentary teeth at some stage in ontogeny (Characters 50).
Premaxillary teeth as a potential synapomorphy: We noted the presence of teeth on the premaxilla of small juvenile specimens (< 50 mm TL) of all three species of Brachyhypopomus for which specimens of this size were available (B. brevirostris, 47.7 mm; B. diazi, 20 mm; B. sp. "palenque", 31 mm), and also in both immature (<75 mm TL) and mature adult (>140 mm TL) specimens of two additional species: B. bennetti and B. walteri. These observations suggest that the presence of premaxillary teeth early in ontogeny may be ubiquitous in the genus. Premaxillary teeth have never been reported for other rhamphichthyoid genera [20,50,51], and we confirmed the absence of teeth on the premaxilla not only in adults of all examined rhamphichthyoid genera outside Brachyhypopomus (i.e. Hypopomus, Hypopygus, Microsternarchus, Racenisia, Rhamphichthys, and Steatogenys) but also in small juvenile specimens (< 45 mm) for Hypopygus, Microsternarchus, Racenisia, and Rhamphichthys. We suspect that a more thorough survey of the osteology of juvenile rhamphichthyoid may reveal that the presence of premaxillary teeth at some stage in ontogeny is a diagnostic character for Brachyhypopomus.
Dentary teeth as a potential synapomorphy: Mago-Leccia [51], Albert & Campos-da-Paz [50], and Albert [20] stated that members of Rhamphichthyoidea are recognized by the absence of teeth on the dentary. However, as with premaxillary teeth, we noted the presence of teeth on the dentary of small juvenile specimens (< 50 mm) of all of three species of Brachyhypopomus for which specimens of this size were available (B. brevirostris, 47.7 mm; B. diazi, 20 mm; B. sp. "palenque", 31 mm), and also in a larger juvenile specimen of B. walteri (73 mm). In contrast, we confirmed the absence of dentary teeth in both adults and small juveniles (< 42 mm) of Hypopygus, Microsternarchus, Rhamphichthys, and Racenisia (note: we were unable to confirm Fernandez-Yépez's [64] report of the presence of dentary teeth in Microsternarchus bilineatus). Based on these observations we propose that the presence of dentary teeth at some stage in ontogeny is a diagnostic character for Brachyhypopomus-although further work is required to confirm this.

Phylogenetic interrelationships within Brachyhypopomus
Below, we discuss phylogenetic relationships among Brachyhypopomus species. We emphasize the results for the total evidence Bayesian phylogenetic analysis, but provide comparisons with the parsimony analysis, and individual gene tree analyses when relevant.
Major clades and species groups. Bayesian total evidence phylogenetic reconstruction (Figs 5 and 7) provides strong support for four major clades within the genus: 1) Clade B comprising the beebei, sp. "belindae", pinnicaudatus, and bennetti species-groups, and B. sp. "flavipomus"; 2) Clade L, comprising the sp. "batesi", bombilla, and occidentalis species-groups, and B. sp. "sullivani"; 3) Clade T-the brevirostris species-group; and 4) Clade W-the janeiroensis species-group. At a higher level, there is strong support for a clade A comprising clades B and L. The internal nodes within the four major clades B, L, T, and W are strongly supported, with only two exceptions: First, a poorly supported Clade 2 comprising the pinnicaudatus species-group and a Clade D comprising the beebei and sp. "belindae" species-groups. Second, a poorly supported clade 3 comprising the bombilla species-group and B. sp. "sullivani". Parsimony total evidence (Fig 6) recovers the same species-groups resolved by Bayesian analysis (Fig 5), and with the same species composition (with the exception of B. sp. "palenque" which appears outside the B. occidentalis species-group in the parsimony tree but is placed within it in the Bayesian tree).
Bayesian total evidence analysis provides weaker support for some of the higher level clades within the genus. For example, poor support for the monophyly of Clade 1 casts uncertainty on the interrelationships between clades A and T, and on the placement of the janeiroensis species-group as sister taxon to the remaining congeners. This instability is reflected by an alternative placement of the janeiroensis species-group in the parsimony topology-as sister taxon to the brevirostris species-group. These higher-level interrelationships will be revised in future analyses that incorporate additional molecular data.
Species monophyly. The topology and high nodal support values reported in the total evidence analysis support the monophyly of all 28 species of Brachyhypopomus, including those with large geographical ranges spanning multiple basins (Fig 18), for which we obtained molecular data from distant sites, e.g. B. bombilla, B. brevirostris, B. gauderio, B. pinnicaudatus, B. occidentalis, and B. walteri. Parsimony analysis resolved B. beebei as a paraphyletic species containing B. sp. "hamiltoni" (Fig 6). However, BI analysis provides strong support (PP = 1) for a monophyletic B. beebei as the sister species of B. sp. "hamiltoni" (Fig 5).
Comparison between phylogenies based on morphology and total evidence. The morphological tree presented in Fig 1 shows several areas of precise congruence with the Bayesian total evidence phylogeny: these include strong support for the bennetti, pinnicaudatus, bombilla, occidentalis, and brevirostris species-groups, and for the Microsternarchini (Microsternarchus bilineatus + Racenisia fimbriipinna)-all represented by precisely the same species. However, there are several areas of incongruence between the morphological and total evidence trees. First, the topologies of early branching in the genus differ, with nodal support for the higher level clades being considerably weaker in the morphology-based tree. Second, the morphological analysis places the Microsternarchini at a deeply nested position within a nonmonophyletic Brachyhypopomus-inside a clade comprising other diminutive, slender-bodied stream-dwelling species from the sp. "batesi" species-group. Third, the morphological analysis does not reconstruct a monophyletic janeiroensis species-group. Finally, the interrelationship of species within Clade B differs from the total evidence analysis, with the exception of the pinnicaudatus and bennetti species-groups.
The nested position of Microsternarchini within the sp. "batesi"-group in the morphologybased phylogeny may be explained as follows. The clade comprising B. sp. "batesi" + B. sp. "provenzanoi" + B. sp. "benjamini" + Microsternarchini exhibits the following unambiguous synapomorphies based on parsimony analysis of morphological characters only: derived loss of first branchiostegal ray (character 19); descending process of maxilla a derived narrow shape, versus broad (character 47). The clade B. sp. "provenzanoi" + B. sp. "benjamini" + Microsternarchini is supported by three additional unambiguous synapomorphies: reversal to ancestral funnel-shape of second basibranchial (character 28); derived loss of gill rakers (character 31); derived loss of scales on dorsal region of anterior third of body (character 45). Further, a clade comprising B. sp. "benjamini" + Microsternarchini is supported by a single unambiguous synapomorphy: derived lack of ossification of the third basibranchial (character 29). A pattern emerges-the synapomorphies supporting the grouping of Microsternarchini with B. sp. "batesi", B. sp. "benjamini" and B. sp. "provenzanoi" (all of which are known to mature at a small body size, less than 80 mm total length, and are confined to small lowland terra firme streams, see Fig 19A) mostly involve derived simplifications of the skeleton and squamation, which we assume to be associated with body size reduction and life in the small interstices of terra firme streams. De Santana & Crampton [49] noted a similar pattern of reductive morphological evolution associated with small body size in Hypopygus, which is also restricted to lowland terra firme streams. Our Bayesian and parsimony total evidence analyses provides strong support for a placing of the Microsternarchini outside Brachyhypopomus. We therefore speculate that the morphological characters uniting the Microsternarchini and the sp. "batesi" species-group of Brachyhypopomus may be homoplastic characters that evolved as convergent responses to similar ecological conditions.
We noted low levels of nodal support in the morphological tree not just for some of the early branching clades (as also observed in the total evidence analysis), but generalized across the phylogeny (except for some very well-supported clades such as the bennetti-group, the sister species B. sp. "cunia" + B. sp. "hendersoni", and the Microsternarchini). Moreover, we were only able to identify 56 parsimony-informative morphological characters, despite the species richness of the genus, and these came mostly from the cephalic region. This apparent paucity of morphological character variation, coupled with a generalized weakness of nodal support may reflect a generalized pattern of morphological trait conservatism within the genus, which we speculate to be the consequence of one or both of two phenomena, described below.
First, morphological trait conservatism is often a consequence of phylogenetic niche conservatism, in which a taxon diversifies within a relatively narrow range of habitats and ecological variables, and consequently with limited adaptive ecomorphological diversification [75]. All species of Brachyhypopomus are restricted to tangled substrates in lentic or slow-flowing environments, and have similar diets of aquatic microinvertebrates, despite some adaptations related to water conductivity and dissolved oxygen availability [2,8].
Second, a pattern of morphological trait conservatism may be an indirect consequence of speciation driven primarily by sexual selection and reproductive character displacement of mate attraction signals carried by the electric organ discharge (EOD). Because the EOD mate attraction signals of gymnotiforms are thought to play a strong role in prezygotic reproductive isolation and speciation [76,77], and because gymnotiform fishes express variation in their electric mate attraction signals based on aspects of electric organ microanatomy and electrophysiology [77,78] that are essentially decoupled from gross external morphology, a mechanism exists for speciation and diversification with little attendant ecomorphological evolution. Arnegard et al. [79] advance similar discussions for the mormyrid electric fish genus Paramormyrops.
The results of our total evidence analyses are also broadly congruent with Carvalho's [22] phylogenetic analysis of a combined morphological and molecular dataset that included 14 species of Brachyhypopomus. Carvalho's tree supports our clades B, C, E, L, and our beebei, pinnicaudatus, and occidentalis species-groups. However, it differs in its basal branching patterns, and in the placing of members of the brevirostris and janeiroensis groups.
Finally, the results of our total evidence analyses correspond partially to a topology for nine species of Brachyhypopomus presented in a phylogeny of all gymnotiforms by Tagliacollo et al [23]. This topology recovered our occidentalis species-group. It also includes a clade comprising the occidentalis species-species + B. sp. "sullivani", listed as B. new sp. 'roy' , (these belong to our clade L), and a clade they resolve comprising B. beebei, B. draco, and B. pinnicaudatus (which belong to our Clade C). Their topology differed in its higher level branching and with regard to the placing of B. brevirostris and B. bullocki.
A comment on subgenera in Brachyhypopomus. Sullivan et al. [63] elected to place B. bennetti and B. walteri within a sub-genus (Odontohypopomus). We do not advocate the use of subgenera to classify clades within Brachyhypopomus and instead recommend the adaptable and taxonomically less burdensome species-group approach that we utilize herein, and that has been applied to other species-rich gymnotiform taxa, e.g. Apteronotus [80] and Gymnotus [77]. Our primary justification for this approach that if a subgeneric name is maintained for B. bennetti and B. walteri, then based on our analyses, at least seven other subgenera might need to be created-corresponding approximately to the species groups annotated in Fig 7. Generic Interrelationships in the Rhamphichthyoidea Although our analyses are designed to focus on species interrelationships within Brachyhypopomus, we found strong support for a clade comprising the hypopomid genera Brachyhypopomus, Hypopomus, and Microsternarchini. Likewise, our analyses provide strong support for a positioning of the Steatogeni (Hypopygus + Steatogenys) as sister taxon to a clade comprising Gymnorhamphichthys + Rhamphichthys. This mirrors the results of Alves-Gomes et al. [21], Arnegard et al. [81], Chen et al. [82], Carvalho [22], and Maldonado-Ocampo et al. [14] and supports the redefinition of the Hypopomidae by Maldonado-Ocampo et al. [14] to comprise Akawaio + Brachyhypopomus + Hypopomus + Microsternarchini and of the Rhamphichthyidae to comprise Gymnorhamphichthys + Iracema + Rhamphichthys + Steatogeni.

Geographical Distributions and Diversification
To model the geographical distributions of Brachyhypopomus in a phylogenetic context, we optimized the distribution of 28 species and three hypopomid outgroup taxa among the five geographical regions described earlier (Results, 'Geographic distributions') (Fig 18). Our analyses unequivocally support an origin of Brachyhypopomus in Greater Amazonia (Region 1)the superbasin comprising the rio Amazonas and río Orinoco basins, and the coastal drainages of the Guianas, sensu Albert & Reis [83]. All hypopomid genera other than Brachyhypopomus (including Akawaio penak and Procerusternarchus pixuna; not included in Fig 18) are entirely restricted to this region, mirroring patterns observed in many other Neotropical fish genera [84], the highest species diversity of Brachyhypopomus occurs in Greater Amazonia, with 21 of 28 Brachyhypopomus species (75%) known from this region (17 exclusively so). From a Greater Amazonian center of origin, Brachyhypopomus has evidently subsequently occupied adjacent systems by a combination of vicariance and dispersal; presumably dispersal across permeable watersheds and geodispersal via river capture events, sensu Albert & Crampton [85].
Like many Neotropical fish taxa of similar taxonomic rank [84], the genus Brachyhypopomus is evidently of considerable antiquity, indicating processes of cross-basin dispersal, speciation, and extinction that have occurred over periods of geological time that in many cases greatly exceed the age of modern drainage boundaries [22]. Chen et al. [82] estimate the Hypopomidae to have diverged from the Rhamphichthyidae (based on the branching of Brachyhypopomus from Steatogenys + Rhamphichthys) ca. 38 Mya (with 95% confidence intervals ranging from the Paleocene to early Miocene). Lavoué et al. [86] dated this event (based on the divergence of Brachyhypopomus from Gymnorhamphichthys) to between the Mid and Late Cretaceous, ca. 100-66 Mya (depending upon fossil calibration protocol), with 95% confidence intervals ranging from the Early Cretaceous to Mid Eocene. Near et al. [87] estimate the divergence of Rhamphichthyoidea from Sternopygidae, an event that must substantially precede the origins of Brachyhypopomus, at ca. 50 Mya.
Below we discuss salient phylogenetic patterns of distribution for each of the five geographical regions in Fig 18. An emergent pattern is that regional assemblages of Brachyhypopomus are polyphyletic in structure, with no evidence of extensive in-situ diversification, including within Greater Amazonia. This pattern, which is perhaps ubiquitous in the continental fish fauna of South America, implies a history of dispersal-assembly-sensu Hubbell [88]-from wider, continental-scale species pools, and over long periods of time [2,85].
We acknowledge that our interpretation of the historical biogeography of Brachyhypopomus is limited by the weak support for clade 1 in our total evidence analyses ( Fig 5). Moreover, because of the antiquity of the genus, distributional patterns have likely been drastically rearranged on multiple occasions, consequently erasing evidence for early vicariance and dispersal. Finally, we recognize that our interpretations are also to some extent contingent on the accuracy of our recognition of widely distributed species with known population-level genetic substructuring (e.g. B. brevirostris), versus closely related species with allopatric distributions (e.g. B. gauderio and B. pinnicaudatus). Region 1. Greater Amazonia. Brachyhypopomus has occupied all major non-shield drainages of Greater Amazonia. In some cases modern drainage divides within Greater Amazonia, or major fall and rapid series correspond approximately to the distributional limits of sister taxa-implying a history of vicariant speciation. Examples include B. sp. "provenzanoi" in the Orinoco, versus B. sp. "batesi" + B. sp. "benjamini" in the rio Negro and Upper Amazon, which may reflect fluvial interconnectance across Amazonian foreland arc prior to a Late Miocene reconfiguration of the Amazon's main tributaries [84,89]. Another example involves three species (B. sp. "alberti", B. sp. "arrayae", and B. sp. "cunia") that are completely, or almost completely restricted to the Upper Madeira, above the major series of cataracts and rapids in its middle course (see also comments for B. bombilla in the discussion of Region 2, below). B. sp. "alberti" + B. sp. "arrayae", and their sister taxon B. draco (confined to Region 2, the La Plata system and Lagoa dos Patos-Merim and associated drainages of SE Brazil) form a sister taxon to B. beebei + B. sp. "hamiltoni", a clade that is absent from the Upper Madeira but is distributed over much of the remaining portions of Greater Amazonia. Likewise, B. sp. "cunia" is found mostly above the middle-Madeira falls (notwithstanding its description from a site just below the first falls), and is sister taxon to B. bullocki + B. sp. "hendersoni", a clade restricted to other parts of Greater Amazonia. Several authors have remarked on the high levels of fish endemicity in the Upper Madeira, and attributed this, in part, to isolation from the lowland Central and Lower Amazon by the middle-Madeira falls-the long series of cataracts and rapids (and the absence of broad riverine floodplains) between Porto Velho and Guajará-Mirim [90][91][92][93]. Other major Amazonian rivers are interrupted by major falls or rapids in their lower or middle courses (e.g. the Xingú, Tapajós, Tocantins, and Negro). We are unaware, as yet, of endemic Brachyhypopomus confined to the headwaters of these systems, but these regions are exceptionally poorly collected.
In other cases, species exhibit distributions that span modern drainage divides within Greater Amazonia, or major waterfalls, suggesting a recent formation of the barriers or relatively recent cross-watershed dispersal. For example, the rio Amazon-río Orinoco divide is bridged by B. bullocki, B. beebei, B. brevirostris, and B. sp. "sullivani"-at least judging from their distributions in both upper and lower Orinoco, as well as the upper and lower Negro. The role of chemical gradients along the río Casiquiare (which connects the Orinoco and Negro) and flanking rapids in the upper rio Negro and río Orinoco as filters for dispersal are discussed by Winemiller et al. [94] and Winemiller & Willis [95] (and see 'Ecological distributions in a phylogenetic context' , below). Other examples of distributions spanning major divides within the Amazon include species that occur both in the rio Negro, and in the Essequibo River (B. bullocki, B. beebei, B. brevirostris, B. sp. "hendersoni", B. sp. "regani", and B. walteri). The Negro and Essequibo exhibit a contemporary connection via the seasonal wetlands of the Rupununi Savanna, at the headwaters of the Essequibo River and rio Branco (a major rio Negro tributary) [96][97][98].
A putative example of dispersal between drainages of Greater Amazonia that are currently completely disconnected involves species that known from the Eastern Amazon and also from small drainages along the Atlantic coastal plain of the Guianas-particularly French Guiana and Suriname. These include B. beebei, B. brevirostris, B. pinnicaudatus, and B. sp. "regani". One of these species, B. pinnicaudatus, exhibits a wide distribution through the Amazon basin, but is absent from the Essequibo (another potential conduit from the Central Amazon via the rio Branco), and thus probably dispersed from the Amazon's modern estuary. Jegú & Keith [99] noted the strong similarity between the freshwater fish population of coastal French Guiana and the main stem of the Amazon River basin, and suggested that the similarity derives from dispersal from the Amazon via its freshwater plume, or by stream capture or interdigitation along the coastal floodplain; see also Lujan & Armbruster [98].
Region 2. La Plata-Lagoa dos Patos. Our total evidence phylogeny suggests that two species, B. draco and B. gauderio, originated from independent dispersal events from the Amazon basin into the Paraguay basin, with subsequent allopatric speciation. In both cases dispersal across the Upper Madeira-Paraguay divide is implicated because the sister taxa are abundant in the Upper rio Madeira (B. sp. "alberti" + B. sp. "arrayae" in the case of B. draco, and B. pinnicaudatus in the case of B. gauderio) but absent from other Amazonian tributaries with headwaters close to Paraguay headwaters (Tapajós, Xingú, and Tocantins-Araguaia drainages).
Hubert & Renno [100], Lovejoy et al. [101], and Carvalho & Albert [93] describe the role of the Amazon-Paraguay Divide as a historically semipermeable barrier-acting not only as a barrier promoting allopatric speciation, but also as a conduit for dispersal, faunal exchanges, distributional range extensions, and secondary contact between previously isolated taxa. The authors also discuss pathways for dispersal via river capture between the Amazon and Paraguay via headwaters of the Mamoré, Guaporé, Tapajós, and Xingú, and comment on the existence of contemporary seasonal wetlands that may permit dispersal across these divides.
Carvalho & Albert [93] note that around one third of the species known from the Paraguay basin also occur in southern Amazon tributaries or other parts of Amazonia, suggesting that a large proportion of species that have crossed the Amazon-Paraguay divide did so recently, and have not yet diverged into diagnosable species; despite a trend for even relatively subtle morphological or genetic variation between populations in species shared between major basins to be assumed to represent species-level divergence [102]. Mirroring this observation for the Paraguay basin ichthyofauna as a whole, three of the five species of Brachyhypopomus known from the rio Paraguay, are also known from southern Amazon drainages: B. bombilla, B. brevirostris, and B. walteri, with the topology of our total evidence analyses implying dispersal into the Paraguay from the Amazon in each case. In contrast to B. brevirostris and B. walteri, whose distributions are mostly centered on the Amazon and other parts of Greater Amazonia, the range of B. bombilla is centered primarily on the rio Paraguay-Paraná-Uruguay, and the Lagoa dos Patos system and adjacent coastal drainages. Populations of B. bombilla from the Upper Madeira are morphologically similar to populations from high southern latitudes, and together these form a monophyletic group in our total evidence analyses. We speculate that B. bombilla may have been isolated from its sister taxon (clade Q, with a distribution encompassing most of Greater Amazonia and the rio São Francisco) by the middle Madeira falls (see above)-with concomitant or subsequent dispersal across the Guaporé-Paraguay Divide. Brachyhypopomus brevirostris and B. walteri are represented in multiple southern Amazon drainages with headwaters abutting rio Paraguay headwaters. Consequently, pathways for dispersal into the rio Paraguay are unknown for these two species but may be elucidated by future population genetic analyses that incorporate populations from the headwaters of the Paraguay, Guaporé, Tapajós, Xingú, and Upper Tocantins-Araguaia. Brachyhypopomus is absent from the entire rio Paraná drainage upstream of the former Guaíra Falls (drowned by the Itaipu hydroelectric dam since 1982), indicating that the colonization of the La Plata drainages by Brachyhypopomus is unlikely to have occurred via an Amazon-Paraná conduit.
The Lagoa dos Patos-Merim drainage, and adjacent coastal drainages: Three species, B. bombilla, B. draco, and B. gauderio exhibit similar southern distributions, which bridge the Paraguay-Paraná-Uruguay system and the Lagoa dos Patos-Mirim drainage (and for B. draco and B. gauderio the rio Tramandaí and rio Maquiné -small coastal drainages adjacent to the Lagoa dos Patos). Extensive faunal sharing between these drainages has been noted in other groups of fishes [103,104].
Region 3. Brazilian coastal drainages. The origins of B. jureiae and B. janeiroensis (which together form a strongly supported B. janeiroensis species-group, clade W), require a separate explanation to the three species from the more southerly coastal Brazilian systems of the Lagoa dos Patos-Merim system and adjacent rio Tramandaí and rio Maquiné (see above). Brachyhypopomus jureiae and B. janeiroensis occur further to the north, and occupy small distributional ranges-B. jureiae from the Ribeira de Iguape [105], and B. janeiroensis from the São João and rio Paraiba do Sul drainages [106]. These limited ranges are relatively unusual for species from these drainages; many other fish groups in the Ribeira de Iguape and Paraiba do Sul systems are commonly also known from upper rio Paraná drainages to the east of the coastal mountain ranges of eastern Brazil [103,107]. Nonetheless, species of Brachyhypopomus are unknown from the upper Paraná.
Our Bayesian total evidence phylogenetic analysis provides some support for a basal divergence between the janeiroensis species-group and all remaining Brachyhypopomus species, which belong to Clade 1. Nonetheless, weak support for the monophyly of Clade 1 (Fig 5), and an alternative placement of the janeiroensis species-group in our parsimony analysis (as sister taxon to the brevirostris species-group, see Fig 6) diminish our confidence in the early branching events in Brachyhypopomus. Regardless of its phylogenetic position, we postulate that the janeiroensis species-group originated by allopatric isolation across the main block of the Brazilian Shield, following colonization of its south-eastern fringe from a Greater Amazonian center of diversification. Similar sister-clades from the Amazon and southeast coastal drainages (without species occurring in intervening areas) have been described in several other Neotropical fish groups; see "pattern B" vicariant patterns described by Ribeiro [103].
Region 4. São Francisco drainage. The rio São Francisco drains a large portion of the central Brazilian Shield but hosts only one species of Brachyhypopomus-B. sp. "menezesi". This species is the only member of the genus that is entirely restricted to a shield drainage. The sister-species relationship between B. sp. "menezesi" and B. sp. "regani", which occurs over wide areas of Greater Amazonia, suggests a history of dispersal from an Amazon drainage and subsequent allopatric speciation, perhaps from a rio Tocantins tributary (B. sp. "regani" is known from the upper Araguaia but not upper Tocantins), or from populations once present in the rio Parnaíba drainage; see Buckup [107] (Brachyhypopomus is not known from the Parnaíba).
Region 5. Trans-Andean drainages. Northwestern South America and the Isthmus of Panama have experienced greater geological upheaval than any other area of the Neotropicsincluding the partitioning of the Maracaibo basin, Magdalena basins, Pacific coastal systems, and Caribbean coastal drainages by multiple Andean orogenies, and the Plio-Pleistocene (or earlier) closure of the Panamanian isthmus. These transformations, and their impact on the ichthyofauna, have been reviewed by multiple authors [69,101,[108][109][110][111]. Two species of Brachyhypopomus, B. occidentalis and B. sp. "palenque", are exclusively trans-Andean. One species, B. diazi occurs in both trans-and cis-Andean drainages.
Our total evidence analyses support a single trans-Andean vicariant speciation event between clade R-the occidentalis species-group and clade M (a clade almost entirely restricted to Greater Amazonia, see Fig 18). This was followed by a divergence between B. sp. "palenque", which occurs in southerly Pacific drainages of Ecuador, and the clade comprising B. occidentalis + B. diazi. Brachyhypopomus occidentalis is known from multiple Pacific and Atlantic drainages of Colombia, Panama, and Venezuela, including the Maracaibo drainage. Brachyhypopomus diazi is the only species of hypopomid that occurs in both cis-Andean drainages (Llanos wetlands of the río Orinoco) and trans-Andean drainages (Caribbean coastal drainages to the north of the Venezuelan Caribbean Cordillera). Our total evidence topology suggests that B. diazi colonized the Orinoco via range extension from trans-Andean drainages, rather than the reverse-despite occupying a wider geographical distribution in the Orinocan Llanos than in the trans-Andean coastal drainages of the Caribbean coast.
We noted a striking degree of genetic similarity between B. diazi populations in the trans-Andean coastal río Salado drainage (the type locality), and in the cis-Andean Llanos. For cytb, all four individuals sequenced across these two areas (Table 3) are identical, except for B. dia-zi_305_OR, which showed a single base pair difference (0.09% uncorrected sequence divergence) from the other three individuals. For rag2, we found a maximum of two base pair differences (0.2% uncorrected divergence) between B. diazi_305_OR and B. diazi_2409_NW. These divergences are consistent with a recent dispersal or translocation event. Dispersal by stream capture across a low-lying area of the Caribbean Cordillera is a possibility, and there are candidate sites for such an event where headwaters of the río Yaracuy (Carribbean drainage) reach within 2-3 km of headwaters of headwaters of the río Portuguesa (Orinoco drainage) at ca. 10°06°N, 068°58'W and at an elevation of around 300 m. Dispersal events from trans-Andean to cis-Andean drainages are not common in Neotropical fish (for a review of cis-trans Andean vicariance see [108]), and B. diazi may represent one of the first cases for which there is strong support from phylogenetic and genetic data.

Ecological Distributions
To model ecological diversification and specialization in Brachyhypopomus in a phylogenetic context, we considered distributions based on habitat occupancy, electrical conductivity, and dissolved oxygen in light of our total evidence BI phylogeny (Figs 19 and 20).
Habitat occupancy. All hypopomid outgroups to Brachyhypopomus are restricted to (lowconductivity) terra firme streams and swamps, or shield stream systems (including Akawaio penak and Procerusternarchus pixuna, not included in our phylogeny). This is reflected by the optimization of habitat and conductivity onto the total evidence phylogeny in Fig 19. Here the ancestral condition for the genus Brachyhypopomus is optimized as terra firme (white) in Fig 19A and low conductivity (white) in Fig 19B with much higher certainty than floodplain (black) or high conductivity (black) or grey (eurytopic for habitat or eurytopic for conductivity). The ancestral habitat of Brachyhypopomus therefore likely resembled a low-conductivity non-floodplain system.
Floodplain specialization has evidently evolved in at least two clades: clade B (both highand low-conductivity floodplains), and clade V-comprising B. sp. "cunia" and B. sp. "hendersoni") (low-conductivity blackwater floodplains only). Within clade B there are subsequent reversals to stream occupation: in B. sp. "alberti" and B. sp. "verdii" (stream specialists) and in B. beebei, B. draco, and B. walteri (eurytopic species). Outside clade B there are some transitions from an ancestral character state optimized with high probability as terra-firme stream to a eurytopic condition-notably B. bombilla, diazi, B. occidentalis, and B. sp. "regani".
Conductivity. The ancestral habitat in Brachyhypopomus is optimized with high probability as low-conductivity. This is mostly retained in clade T, with a transition to a eurytopic condition in B. brevirostris. Although the ancestral character states of higher level clades 1, A, and B are ambiguous, a pattern emerges in which species in Clade B are mostly specialized to high conductivity systems, or eurytopic-with reversals to the occupation of low-conductivity systems in B. sp. "alberti", and B. sp. "verdii". Within Clade L, the cis-Andean species in Clade M are mostly specialized to low-conductivity systems, with some derived transitions to eurytopy. In contrast, the trans-Andean species belonging to clade R (the occidentalis species-group) optimize as high-conductivity specialists, with a transition to eurytopy in B. occidentalis.
Occurrence in floodplains (Fig 19A) correlates approximately to occurrence in high-conductivity systems (Fig 19B), and likewise occurrence in terra-firme streams correlates approximately to occurrence in low-conductivity systems. These correlations occur because whitewater floodplains and lowland terra firme streams (in which Brachyhypopomus are especially diverse) are characterized by high and low conductivity, respectively. However, the correspondence is imperfect because the Andean and Panamanian piedmont terra firme streams inhabited by B. diazi, B. occidentalis, and B. sp. "palenque" are characterized by high-conductivity. Likewise, the blackwater floodplain systems inhabited by members of the brevirostris species-group are characterized by low conductivity (see 'Ecological distributions', in Results).
Models of impedance matching presented by Hopkins [112] predict correlations between conductivity and the arrangement of electrocytes in the caudal portion of the electric organ, which is located mainly in the caudal filament and generates the high amplitude component of the EOD used in communication [113]. In low conductivity systems, maximum EOD power is associated with a predominantly serial configuration of the electrocytes in a long caudal filament, while in high conductivity systems power is maximized by a parallel configuration of electrocytes in a short caudal filament. As predicted by these models, Brachyhypopomus specialists of high conductivity systems have short tails with a parallel arrangement of electrocyte (e.g. B. bennetti, B. diazi, B. occidentalis, B. sp. "palenque"), while species specialized to low conductivity systems have relatively long tails (e.g. B. brevirostris, B. bullocki, B. sp. "cunia", B. sp. "hendersoni", B. janeiroensis, B. jureiae). These characters are also known to be exaggerated in the breeding males of some of these species [5,63,71,112]. The optimization of conductivity on our total evidence tree ( Fig 19B) indicates that salient impedance matching adaptations to high conductivity may have evolved in response to a transition from an ancestral low-conductivity system independently in at least two lineages: in Clade B (B. bennetti and B. beebei), and in all three species in the occidentalis species-group. Nonetheless, some species endemic to high conductivity systems (B. sp. "arrayae", B sp. "flavipomus"), and to low conductivity systems (B. sp. "alberti", B. sp. "batesi", B. sp. "benjamini", B. sp. "menezesi", B. sp. "provenzanoi". B. sp. "sullivani") exhibit no obvious electric organ or caudal filament specializations of the kind associated with impedance matching, and no salient sexual dimorphism. Impedance matching to a relatively narrow range of conductivity has been predicted to act as a barrier to the dispersal of Brachyhypopomus species, given the geographical distribution of low and high-conductivity rivers in the Neotropics [71,112]. The geographical distributions and habitat occupancies summarized in Figs 18 and 19 provide some support for this prediction. For example, low-conductivity blackwater conditions are found across the Orinoco-Negro divide, and Winemiller & Willis [95] have argued that this allows the dispersal of blackwater adapted fishes or those tolerant of variable water conditions. Brachyhypopomus species bridging the Orinoco-Negro Divide (labeled OR and RN in Fig 18) are, as predicted, all either low-conductivity blackwater specialists (B. bullocki and B. sp. "sullivani") or eurytopic (B. beebei and B. brevirostris). Likewise, species shared between the Essequibo (part of the GU region listed in Fig 18) and rio Negro (RN), which are both low-conductivity blackwater systems, are also either low-conductivity blackwater specialists (B. bullocki, B. sp. "hendersoni"), or eurytopic (B. sp. "regani", and B. walteri). In contrast, species that are specialized to high-conductivity whitewater systems are evidently unable to traverse long corridors of low-conductivity water (often with rapids) to reach similar habitats in adjacent basins (e.g. B. diazi, restricted to the Orinoco basin, and B. bennetti, B. sp. "belindae", B. sp. "flavipomus", and B. pinnicaudatus, restricted to the Amazon basin).
Hypoxia. The capacity to occupy habitats with prolonged hypoxia (Fig 20) approximately mirrors the occupation of high-conductivity whitewater floodplains ( Fig 19A). All clade B members, except B. sp. "alberti" and B. sp. "verdii", occur in hypoxic habitats, as do several species in clade P and R. The occupation of hypoxic habitats has also arisen sporadically in three other species: in B. brevirostris, and also in B. janeiroensis, and Racenisia fimbriipinna (both of which are known to occur in hypoxic terra firme swamp habitats). The ancestral condition regarding tolerance of hypoxia is ambiguous.
Most species of Brachyhypopomus persist in hypoxic water by undertaking aerial gill respiration-either by aspirating bubbles of air into their gill chambers, or by opening their mouths at the surface to expose the gills to air (B. beebei, B. gauderio, B. sp. "flavipomus", B. pinnicaudatus, and B. walteri) [72,114,115]. Crampton et al. [116] noted that two species endemic to seasonally anoxic whitewater floodplains (B. bennetti and B. sp. "flavipomus") have significantly larger gills than two species endemic to normoxic terra firme streams (B. sp. "sullivani") or blackwater floodplains (B. sp. "hendersoni"). Brachyhypopomus species also exhibit a reduction of motor and EOD activity under hypoxic conditions, presumably to save metabolic energy [72].
Ecologicallycosmopolitan species and geographic ranges. We observed a tendency for ecologically eurytopic species (which are tolerant of a range of conductivity, dissolved oxygen, and other conditions) to occupy wider geographical ranges than stenotopic species-matching observations for other gymnotiform species, e.g. Gymnotus carapo, and Sternopygus macrurus, and other Neotropical fish taxa [22]. For example, several species of Brachyhypopomus in Greater Amazonia exhibit ecological distributions that include both high-conductivity, seasonally-hypoxic floodplain systems and low-conductivity, perennially normoxic blackwater or clearwater systems (Figs 19 and 20): B. beebei, B. brevirostris, B. sp. "hamiltoni", B. sp. "regani", and B. walteri. With the exception of B. sp. "hamiltoni", these species exhibit the widest geographical distributions known among congeners.
Albert et al. [53] and de Santana & Vari [29] seek explanations for why Gymnotus and Sternarchorhynchus, respectively, exhibit elevated species richness. The hypotheses advanced for high species richness in Sternarchorhynchus are not applicable to Brachyhypopomus. Key ecomorphological innovations such as the grasp-suction feeding mode of Sternarchorhynchus, which is postulated to have triggered an adaptive radiation in the genus [29], are lacking in Brachyhypopomus. Sternarchorhynchus is also inferred to have diversified in part through allopatric speciation in fast-flowing shield headwater systems located above cataracts, but Brachyhypopomus is absent or uncommon in these systems.
The reasons for high diversity in Brachyhypopomus appear to more closely parallel those for Gymnotus. These include a wide geographical range encompassing multiple drainages outside Greater Amazonia (trans-Andean systems included), the occupation of multiple shallow-water habitats, tolerance of hypoxia (which permits the occupation of whitewater river floodplains and terra firme swamps), and high interspecific EOD diversity [2,53]; data from Gymnotus suggest that electric communication signals may play an important role in reproductive isolation and speciation [5,76,120]. As in Gymnotus, Brachyhypopomus generate a diversity of EOD signals-with sympatric species usually exhibiting distinct, species-typical waveforms and/ or repetition rates [5,71,[121][122][123][124][125][126][127]. The phylogenetic framework established herein makes the genus Brachyhypopomus a superb model system for future investigations of communication signal evolution and species diversification.