Three new species of the Eigenmannia trilineata species group (Gymnotiformes: Sternopygidae) from northwestern South America

Eigenmannia is one of the more taxonomically complex genera within the Gymnotiformes. Here we adopt an integrative taxonomic approach, combining osteology, COI gene sequences, and geometric morphometrics to describe three new species belonging to the E. trilineata species group from Colombian trans-Andean region. These new species increase the number of species in the E. trilineata complex to 18 and the number of species in the genus to 25. The distribution range of the E. trilineata species group is expanded to include parts of northwestern South America and southern Central America.


INTRODUCTION
Eigenmannia Jordan, Evermann, 1896 is the most diverse genus in the neotropical electric fish family Sternopygidae, with 22 valid species (Dutra et al., 2014; Tab. S1 -available only in the online version). The genus occupies trans-Andean drainages of Panamá and northwestern South America and is widely distributed in cis-Andean drainages to as far south as northern Argentina (Albert, 2001). Eigenmannia is considered one of the most taxonomically confused genera in the order Gymnotiformes (Dutra, 2015), although phylogenetic relationships within Eigenmannia have received some recent attention (Tagliacollo et al., 2016;. Peixoto et al. (2015) proposed an Eigenmannia trilineata López, Castello, 1966 species group comprising 15 species based on a single exclusive synapomorphy -the presence of a superior medial stripe on the flank. The E. trilineata species group has not previously been reported in trans-Andean basins (drainages along the Western slope of the Eastern cordillera that flow into both the Pacific and the Caribbean). However, three Eigenmannia species outside this species group are known from this region: E. humboldtii (Steindachner, 1878), E. meeki Dutra, Santana, Wosiacki, 2017, andE. virescens (Valenciennes, 1836) (e.g., Maldonado-Ocampo, Matamoros et al., 2015;Dutra et al., 2017). Recently, Peixoto et al. (2015) restricted E. virescens to forms without dark horizontal stripes from the Paraná river and de La Plata river basins. This range-restriction invalidates records of Eigenmannia ascribed to E. virescens from the Catatumbo, Ranchería, Magdalena, Sinú, Atrato, Tuira, and Bayano rivers Albert, Crampton, 2006;Maldonado-Ocampo, 2011). Because all trans-Andean forms previously ascribed to E. virescens present a superior medial stripe on the flank, we instead propose that they belong to the Eigenmannia trilineata species group.
In this study, we adopted an integrative taxonomic approach (Schlick-Steiner et al., 2010) to reveal and describe three new species of Eigenmannia belonging to the E. trilineata species group from trans-Andean drainages of northwestern South America and southern Central America. We used data from osteology, morphometrics, and COI gene sequences for species delimitations, following a similar recent approach (Santana et al., 2019). These new species increase the number of Eigenmannia species to 26 (here E. goajira Schultz, 1949, which was assigned incertae sedis to the Eigenmanniinae, Dutra et al., 2014, is excluded from the species count) and the number of species in the E. trilineata species group to 18 (Dutra et al., 2014;. The three new species described herein also expand the range of distribution of the E. trilineata species into trans-Andean drainages of northwestern South America and parts of southern Central America.
Semi-landmarks followed the trajectory of three curves in the individual: (1) Dorsal head profile (15 semi-landmarks from landmark 1 to 6), (2) Gill opening (9 semilandmarks from landmark 8 to 9), and (3) Ventral profile of the head (12 semi-landmarks from landmark 14 to 13), (Fig. 1). Type 1 landmarks (intersections of structures, curves, or center of structures) and type 2 landmarks (sections of maximum curvature) were positioned at the start and endpoints of commonly taken measurements on Gymnotiform fishes (Mago-Leccia, 1978;Hulen, 2005;Zelditch et al., 2012). Caudal landmarks were avoided due to variable degrees of caudal filament damage and/or regeneration, and due to post-fixation twisting of the posterior body portions of some specimens.
Subsequently errors were verified on the digitization of landmarks, for which a Principal Component Analysis (PCA) of 10 photos of the same individual in the IMP package (Sheets, 2004) was carried out. When no errors were found, a Generalized Procrustes Analysis (GPA) was performed, using the torsion energy method to define the sliding of the semi-landmarks. The centroid size (CS), the square root of the sum of the squared distances between the landmarks and the centroid configuration, were extracted to perform tests such as allometric analysis and homogeneity of allometric slopes. After eliminating the allometric effects, a between-group PCA (BGPCA) was performed. This allowed the main components of morphological variation to be exported to tangent space, thus detecting the morphometric variation of the samples. A Randomized Residual Permutational Multivariate Analysis of Variance/Canonical Variate Analysis (PERMANOVA/CVA) was then carried out to detect the effect of the aggrupation in the morphometrical differences. Finally, a pairwise test between the least-squares means of each group was performed using the Goodall's F-test, with 1000 bootstrap replicates.
Morphometric measurements and meristic. Linear measurements were taken from the same individuals and groups used in geometric morphometrics, also from their left side. All measurements, counts, and notations follow Mago-Leccia (1978) and Albert (2001).

DNA Extraction and Sequencing.
For the use of the DNA Barcode (COI), muscle tissue samples were obtained from individuals from the basins listed in Tab. S2 (Tab. S2 -available only in the online version). The DNA extraction was performed using 20 mg of muscle tissue in the Qiagen BioSprint 96 extraction robot at the Smithsonian Tropical Research Institute (STRI). DNA was amplified using COI universal primers (Baldwin et al., 2009; COI Fish-BCL: 5' TCAACYAATCAYAAAGATATYGGCAC3', and COI Fish-BCH: 5' ACTTCYGGGTGRCCRAARAATCA3'). For sequencing, the samples were sent to Macrogen. Additionally, COI sequences obtained by Maldonado-Ocampo (2011) and sequences deposited on both GenBank and BOLD assigned to E. virescens or Eigenmannia sp. collected on trans-Andean basins, were used in the analysis.
Sequence Analysis. Sequences were aligned using ClustalW in Bioedit v7.0.0 (Hall, 2011); a FASTA file was generated with all sequences. The substitution saturation was verified in DAMBE (Xia, Xie, 2001), and the result showed a value close to 1. The best nucleotide evolution model for the COI gene was specified via the Akaike Information Criterion using JmodelTest 2.1.7 (Darriba et al., 2012).

Species delimitation.
The generalized mixed Yule coalescent (GMYC) analysis was used to delineate species using genetic data (Zhang et al., 2013), which is based on a prediction that leads to the appearance of distinct genetic clusters separated by longer internal branches (Pons et al., 2006;Fujisawa, Barraclough, 2013). To perform the GMYC analysis, BEAUti software was used to create the .xml file containing the features (e.g., partitions, sites, clocks, trees, MCMC) of the analysis carried out in BEAST v1.8.3 (Drummond, Rambaut, 2007) from the nucleotide alignment in Nexus format. A relaxed lognormal molecular clock tree was estimated with a birth-death model, which resulted in an ultrametric Bayesian tree based on the COI gene. The ESS values and the log files were evaluated in Tracer v1.6.0 (Rambaut et al., 2014). The maximum clade credibility tree was then obtained from the combined trials, after removing 20% of the trees in a "burn-in" in TreeAnnotator v1.8.2 (Rambaut, Drummond, 2016). Finally, the Newick tree was implemented in the splits 1.0 package (Fujisawa, Barraclough, 2013) in R, to obtain the threshold on the submitted tree (GMYC; Fujisawa, Barraclough, 2013). Second, the ABGD model, where the Barcode gap is used as a threshold to delimit species (Puillandre et al., 2012). Sequence alignments were loaded onto the ABGD website at http://wwwabi.snv.jussieu.fr/public/abgd/abgdweb.html and executed with the default settings (Pmin = 0.001, Pmax = 0.1, Steps = 10, X (relative gap width) = 1.5, Nb bins = 20), with the Kimura model (K80) selected for distance. Finally, only the main partitions obtained through the model were utilized. Lastly, genetic distances were calculated based on the Kimura 2-Parameter (K2P) nucleotide substitution model using the program MEGA 7 (Kumar et al., 2016).

Geometric Morphometrics.
A total of 710 individuals were utilized for photographs and procedures. Individuals were organized into eight different groups corresponding to eight drainage basins (Tab. 1). A cis-Andean group comprising individuals belonging to Amazonas, Orinoco, and Paraná basins was included for comparison. PERMANOVA/CVA showed a highly significant (p<0.001) effect of size on the shape, thus demonstrating the expected allometric effect on the data (Tab. 2).
After detecting the allometric effect on the data, it was necessary to verify if the individuals within each of the groups defined for this analysis had similar or different allometric slopes, accounting for different allometric effects within each group. The homogeneity of slopes test revealed that the null hypothesis of homogenous (i.e., parallel) slopes, among groups, was accepted (Tab. 3), thus avoiding the need to correct allometry differentially for each group and instead of complete allometric correction of the full dataset (Fig. 2). Size-adjusted residuals and allometry-free shapes were obtained as the allometric-corrected data. A re-run of the PERMANOVA/CVA no longer showed a significant effect of size on shape (Tab. 4).
The first BGPC accounted for 75.7% of the total shape variation in the dataset, while the second BGPC accounted for only 7.37% of total shape variation (Tab. 5). The high variation represented by the BGPC1 was mainly influenced by the length differences found at the anal-fin origin relative to the snout tip and the opercular opening. These observations suggest that the position of the anal-fin origin has taxonomic importance -especially in discriminating trans-Andean from cis-Andean taxa (yellow in Fig. 3); the origin of the anal fin appears much closer to the opercular opening (almost under it) in the cis-Andean taxa than in trans-Andean taxa (Fig. 3). The second BGPC is much more conservative about changes along the fish body. Maximum values on the BGPC2 correspond to a broader cephalic region and a distant position of the lower operculum relative to the mouth area (Fig. 3).
PERMANOVA/CVA for differences between groups showed highly significant differences between almost every pair of groups, except for between Upper Magdalena and Atrato, where the differences were not shown to be significant, indicating

Species Delimitation.
Genetic distances between the GMYC-derived groups were in all cases higher than 2% except for the difference between the Catatumbo group and the Maracaibo group, where the genetic distance was lower. The genetic distances between ABGD-derived groups were in all cases higher than 2%, in some cases reaching up to 14%. The ABGD analysis found a single group uniting Catatumbo and Maracaibo specimens. Additionally, genetic distances inferred with the K2P also support the results   Taxonomy. Based on the evidence described above, four new species for the trans-Andean region are proposed for the genus Eigenmannia, three of which are described below. All four species belong unambiguously to the Eigenmannia trilineata species group defined by Peixoto et al. (2015) because they all present the character of a superior medial stripe on the flank between the lateral line and the anal-fin base stripe. The fourth species, from the upper Atrato river basin, will be described in a subsequent publication. Description. Morphometric and meristic data is shown in Tab. 8. Total Length (mm): 157.0-220.8 mm. Ovoid scapular foramen. 68-79 vertebrae. 13 precaudal vertebrae. 27 premaxillary teeth in 3-4 rows. 9-10 endopterygoid teeth in 1-2 rows. 20-22 teeth in dentary in 2 rows. Parietal and frontal bones articulated by synchondrosis by means of the frontoparietal suture (in this case the character is noteworthy the edges of the bones where they articulate have deep endings of variable size, but with a length greater than the teeth). Four or more foramina on the opercular opening of the frontal bone. Posterior-most arm of the lateral ethmoid bone elongated; articulates with the most anterior prolongation of the frontal bone forming a window. Width of anterior region of parasphenoid bone equal to width of two premaxillaries. Parasphenoid compressed towards posterior region (giving appearance of a glass bottle). Branchiostegal rays 5 in total, 4 is the largest. Urohyal, laminar and convex until middle par (from here it forms the urohial process, which extends to end of urohyal) Operculum upper edge rounded, bottom edge slightly extended. Endopterygoid with long ascendant process. Basihyal length ca. 1/2 of length of first ceratobranchial, slightly wider in anterior region, almost rectangular. Five ceratobranchials with width conserved. Five basibranchials of which 2-3 are well ossified. Four hypobranchials, first 3 well ossified. Tooth plate of the upper pharyngeal with 7 teeth. Tooth plate of the lower pharyngeal with 12 teeth. 10 cartilaginous gill rakers. Coloration in alcohol. Background color pale yellow to dark orange. Head brown/ dark dorsally and rapidly becoming lighter through the snout. Lips and suborbital region with brown chromatophores slightly darkening the area. Body with four visible dark horizontal stripes: (1) Lateral-line stripe narrow and dark, one scale deep, from the first perforated scale to the caudal filament; (2) Superior medial stripe wide, formed by often scattered spots, extending from the mid-portion of the gas bladder to approximately half of the anal fin; (3) Inferior medial stripe dark and wide, almost two scales deep, well colored and darker in region of the caudal filament, generally starting near the anus and continuing until the end of the anal fin; (4) Stripe along anal-fin base dark, one and a half scales deep, extending along the base of the anal fin. Pectoral fin with scattered dark chromatophores generally near the base and anal fin hyaline. Dark humeral spot marking the beginning of the lateral-line stripe. Nape dark.

Non
Geographic distribution. Eigenmannia camposi is known from the lower Atrato basin, and in the upper and middle Magdalena and Cauca river basins in Colombia (Fig. 7). Diagnosis. Eigenmannia magoi can be distinguished from its congeners of the E. trilineata species group, except E. trilineata and E. zenuensis by the number of premaxillary teeth: 32 in 4 rows (vs. 8-10 in 2 rows in E. muirapinima; 8-12 in 2 rows in E. antonioi; 9-10 in 2 rows in E. guairaca; 11-15 in 3 rows in E. loretana; 13-16 in 3 rows in E. pavulagem; 16 in 3 rows in E. microstomus; 17 in 3 rows in E. sayona; 17-20 in 3 rows in E. correntes; 18-29 in 3-4 rows in E. besouro; 22-24 in 4 rows in E. matintapereira; 24-25 in 4 rows in E. desantanai; 25-26 in 4 rows in E. vicentespelaea; 27 in 3-4 rows in E. camposi and 35-40 in 5 rows in E. waiwai); it can further be distinguished from E. trilineata and E. zenuensis by the number of teeth in the dentary: 35-39 in 2-3 rows (vs. 23 in 2 rows in E. trilineata and 56-60 in 4-5 rows in E. zenuensis); finally it can further be distinguished from E. zenuensis by the shape of the vomer with extensions separated at the posterior end (vs. vomer extensions forming an inverted ridge in E. zenuensis) and by the lower number of teeth in the upper pharyngeal plate 5-6 (vs. 7 in E. zenuensis).
Description. Morphometric and meristic data is presented in Tab. 10. Total Length (mm): 158.0-298.9. Ovoid scapular foramen. 74 vertebrae; 14 precaudal vertebrae. 32 premaxillary teeth in 4 rows. 11 endopterygoid teeth in 2 rows. 35-39 teeth in dentary in 2-3 rows. Suture between parietal and frontal bones serrated with variable teeth size. Three or more foramina on the opercular opening of the frontal bone. Posterior-most arm of the lateral ethmoid bone elongated, articulates with anterior-most prolongation of frontal bone forming window. Orbitosphenoid bone wide. Pterosphenoid bone wide. Posterior edge of pterotic bone not rounded, instead oval and extended. Branchiostegal rays 5 in total, 1-2 of similar length, 3-5 with wide extensions, branchiosteal 4 the largest. Operculum upper edge rounded. Endopterygoid with long ascendant process. Basihyal length approximately 1/2 of length of first ceratobranchial, slightly wider in anterior region, almost rectangular. In all ceratobranchials width is conserved. 4 basibranchials, 1-2 well ossified, 3 lightly ossified. Four hypobranchials, first 3 well ossified. Tooth plate of the upper pharyngeal with 5-6 teeth. Tooth of the lower pharyngeal plate with 12 teeth. 10 cartilaginous gill rakers.

Coloration in alcohol.
Background color pales white to pale yellow. Head dark, darker on the dorsal region getting lighter gradually to the ventral region. Upper lip usually darker than the lower lip, but generally slightly dark; suborbital region dark with concentrated chromatophores. Body with four visible dark horizontal stripes. Above the lateral-line stripe, dispersed chromatophores present along myomeres, making them more observable. Lateral-line stripe thin narrow, one scale deep, the first half portion of the stripe light in color in the anterior half of body, darker on posterior half; stripe starts on the first perforated scale and extends posteriorly to base of the caudal filament. Superior medial stripe very light-colored, made up of dispersed chromatophores, more visible on posterior two-thirds of the body. Inferior medial stripe dark and thick wide, two scales thick wide, originating above anus and extending along length of the anal fin. Stripe along anal-fin base thin narrow, one scale thick wide, dark, extending along the base of the complete entire anal fin. Pectoral fin slightly dark-colored, mainly hyaline, darker near its base. Anal fin hyaline. Humeral spot present, dark brownish. Nape dark.
Geographic distribution. Eigenmannia magoi is known only from the Catatumbo river basin, which drains into Lake Maracaibo in Colombia and Venezuela (Fig. 7).
Etymology. The specific epithet "magoi" is assigned to the new species in honor of Francisco Mago Leccia (1931Leccia ( -2004, for his contributions to our knowledge of gymnotiform fishes.

DISCUSSION
Molecular and morphological data have questioned the monophyly of Eigenmannia (e.g., Alves-Gomes, 1998;Albert, 2001;Maldonado-Ocampo 2011). Dutra (2015), including in their analysis of eight valid Eigenmannia species and eight undescribed species (some of them currently described within the E. trilineata species group,) proposed two synapomorphies to support the monophyly of Eigenmannia: (1). The extension of the epipleural myorhabdoi bones over vertebrae 7-9; (2) The presence of an obliquus inferioris muscle limiting the posteroventral margin of the muscular hiatus between the first rib and the second rib.
However, as described by Galindo-Cuervo (2019), the two characters described above from Dutra (2015) are not exclusive to Eigenmannia, insomuch as these two characters also occur in Sternopygus (Fig. 10). Moreover, we noted that the degree to which the epipleural myorhabdoi bones extend over vertebrae 7-9 varies between species in both Eigenmannia and Sternopygus. For example, in E. camposi the epipleural myorhabdoi bones extend until vertebrae 6 and in E. humboldtii until vertebrae 9 (Fig. 10). Also, the obliquus inferioris muscle limiting the posteroventral margin of muscular hiatus is between the fifth and sixth rib in E. camposi and between the fourth and fifth rib in E. humboldtii (Fig. 10). Based on these observations, Eigenmannia evidently still lacks unambiguous diagnostic characters common to all congeners. Consequently, the three new species described herein are provisionally assigned as members of Eigenmannia based on the absence of almost all diagnostic characters proposed for the remaining genera within Eigenmanniinae.
The three new species described herein are included in the E. trilineata species group based on the presence of the superior medial stripe on the flank, a character considered to be a synapomorphy supporting this species group (Peixoto et al., 2015). Nonetheless, we acknowledge that the monophyly of the E. trilineata species group, as proposed by Peixoto et al. (2015), was not recovered in two phylogenetic studies that included some of the species of this species group: Dutra (2015)