Taxon sampling

We generated UCE data from 95 samples representing eight of the nine valid species of trans-Andean Hyphessobrycon (Fig 1 and S1 Table) and nine species of cis-Andean Hyphessobrycon. Additionally, we included nine characin species as outgroups (S1 Table). Our samples were either collected by the authors or acquired through museum loans from AUM, IMCN, FMNH, LSUMZ, SLU, STRI, UCR, MZ-UNICACH, and UARC-IC. Acronyms follow Sabaj 2020 [35]. Field collected specimens were euthanized following an approved IACU protocol at Louisiana State University (IACUCAM-21-103). Fin clips were stored in 95% ethanol and voucher specimens were preserved in 10% formalin, stored in 70% ethanol, and deposited in the fish collections of LSUMZ, FMNH, and SLU. Research and collecting permits (DRM/002/2018; 48–2018) in Guatemala were granted by Consejo Nacional of Areas Protegidas (CONAP) to DJE.

DNA extraction and amplification of mitochondrial genes

We extracted genomic DNA from a total of 113 samples of characin species using DNeasy spin columns (QIAGEN). We assessed DNA quality and quantity using 1% agarose gels, and a Qubit® 2.0 fluorometer (Invitrogen). To corroborate the taxonomic identity of our samples, we amplified and sequenced the partial Cytochrome Oxidase I (COI) “barcode” gene using the primers and protocol of Ward et al. [36] and compared the sequences against the NCBI database using the nucleotide basic local alignment search tool—blastn—online (https://blast.ncbi.nlm.nih.gov/Blast.cgi) [37].

UCE library preparation

We used ~300 ng of DNA as starting template and sheared the DNA to a target fragment size of 400–600 base pairs (bp) [38] using an EpiSonic Multi-functional bioprocessor (EpicGenTek). We used the Kapa Hyper Prep Kit (Kapa Biosystems) to construct dual index libraries. Libraries were pooled and enriched following protocols in https://www.ultraconserved.org/#protocols with adjustments of Burress et al. [39]. We then used the myBaits UCE Ostariophysan 2.7Kv1 kit (Daicel Arbor Biosciences) to target the capture of 2,708 UCE loci from our samples [40]. Libraries were hybridized and enriched in pools of eight samples. Prior to sequencing, we performed quality control of the final libraries using Bioanalyzer (Agilent). Finally, we combined all libraries in a final pool with a concentration of 10μM, which was then sent for sequencing on one lane of PE150 Illumina HiSeq at Novogene (https://en.novogene.com/). The generated raw reads are archived in the sequence repository archive (SRA) of the National Center for Biotechnology Information (NCBI) (BioProject PRJNA887072)

We also obtained raw sequences from 87 individuals included in a recently published phylogeny of characoid fishes [9] (SRA BioProject PRJNA563917) that belong to seven subfamilies of the family Characidae: Stethaprioninae (63 species), Stevardiinae (six species), Cheirodontinae (three species), Characinae (two species), Exodontinae (two species), Tetragonopterinae (two species), Aphyocharacinae (two species); and Spintherobolinae (four species) (S1 Table). Therefore, our final dataset for analysis included a total of 108 species of Characidae, including 20 species of Hyphessobrycon representing eight trans-Andean species and 12 cis-Andean species that included five species assigned to the ‘rosy tetra’ clade sensu Weitzman and Palmer [32].

Bioinformatics

We used the software PHYLUCE v1.7–1 [41] to perform bioinformatic processing of sequencing reads, identification and extraction of UCE loci from reads, and alignment of UCE loci across samples for downstream phylogenomic inference. Adapter contamination and low-quality reads were removed from our newly generated demultiplexed raw reads using Illumiprocessor [42], a wrapper for Trimmomatic [43] implemented in the PHYLUCE pipeline using the command illuminoprocessor. For the Melo data we used fastp [44] implemented in YAP v0.2.1 (Yet-Another-Pipeline) [45] to perform the cleaning of this dataset. The input file for fastp was generated using the command new -d and then we used the command qc -i to clean the raw data. Finally, we combined all the clean reads (newly generated + Melo data; FASTA files) into a single folder to continue processing the data.

We assembled the data for each individual sample into contigs using SPAdes [46, 47] within PHYLUCE using the program phyluce_assembly_assemblo_spades. To identify and extract UCEs we ran the assembled contigs for all the samples against the ostariophysan probe set (2,708 UCEs) [40] using the program phyluce_assembly_match_contigs_to_probes. Then we generated a list of all loci for all samples using the program phyluce_assembly_get_match_counts that was used to extract the FASTA data using the program phyluce_get_fastas_from_match_counts. FASTA files were exploded using the program phyluce_assembly_explode_get_fastas_file to obtain summary statistics for each sample assembled. We aligned the UCEs using the MAFFT v.7 aligner [48], first with edge trimming followed by internal trimming using the program phyluce_seqcap_align. Subsequently, we trimmed locus alignments using Gblocks [49] using the program phyluce_align_get_gblocks_trimmed_alignments_from_untrimmed. Trimmed alignments were cleaned to remove the locus name and just keep the sample name using the program phyluce_align_remove_locus_name_from_files. Finally, we generated three data sets that varied in the level of completeness of each alignment (viz, 75, 90, and 95%) using phyluce_align_get_only_loci_with_min_taxa and calculated summary statistics for each alignment using SEGUL [50]. We used these datasets in downstream phylogenomic analyses. All the data processing steps described above follow standard protocols (https://phyluce.readthedocs.io/) and were implemented on the Louisiana State University High Performance Computing Cluster SuperMIC.

Phylogenomic inference

Phylogenomic hypotheses based on the concatenated alignments were inferred for all three datasets (i.e., 75, 90, and 95%) under a maximum likelihood approach implemented in IQ-TREE 2 [51]. Model selection for each UCE locus was performed using ModelFinder [52] implemented in IQ-TREE 2 followed by tree inference in the same run using the option -m TEST. Node support was evaluated using both Ultrafast boostrap (UFBoot2) [52, 53] and the SH-like approximate ratio test (SH-aLRT) [54] with 1000 replicates each. Finally, we estimated gene trees and their support values (UFBoot2; 1000 replicates) for all loci for the three datasets (75, 90, and 95%) in IQ-TREE 2 using the options -s and -bb.

We inferred a species tree using ASTRAL-III, a two-step method [i.e., inference of gene trees (step 1) and species tree (step 2)] that is consistent with the multispecies coalescent model [55]. ASTRAL-III uses gene trees as input for the inference of the species tree. To minimize the effects of gene tree estimation error, see Simmons and Gatesy [56], we collapsed nodes with values < = 50 bootstrap support (i.e., UFBoot2) in all gene trees in Newick utilities [57] using the option nw_ed b < = 50 prior to inferring the species tree. These “collapsed” gene trees were used as input to infer the species tree and to calculate local posterior probabilities (LPP) in ASTRAL-III. In order to have one species or lineage per tip represented in our inferred species tree, we mapped terminal branches of the gene trees representing the same species or lineage into the same taxon (i.e., terminal branch of the species tree) using the option -a in Astral-III.

Recently, it has been highlighted that standard measures of support in phylogenomic inference (i.e., bootstrap support and Bayesian posterior probabilities) have some limitations [58, 59]. An expectation is that these measures of support will reach their maximum values even in cases of high gene tree discordance (see Thomson and Brown [59]) possibly leading to high support in topological hypotheses that might not accurately reflect the evolutionary history of the group of interest. We utilized gene concordance factors (gCF) [51, 60] and normalized quartet scores (NQS) [61] as alternative measures of support of relationships in our concatenated and coalescent phylogenomic hypotheses respectively. We used the collapsed gene trees as input to calculate gene concordance factors (gCF), the frequency of the two alternative topologies (i.e., gene discordance factors; gDF1 and gDF2), and the proportion of gene trees discordant due to polyphyly (gDFP; see Mihn et al. [51]) in our concatenated topology in IQ-TREE 2 using the options—gcf and—df-tree. We annotated our species tree using the option -t 16 in ASTRAL-III to calculate the NQS of the main topology (preferred species tree) and its branches and the two alternative topologies. For both of these analyses we focused on the recovered relationships of trans-Andean species of Hyphessobrycon within Stethaprioninae (monophyly vs non-monophyly). Furthermore, we investigated the recovered relationships between species of trans-Andean Hyphessobrycon.

Investigation of cryptic diversity

We investigated if there is cryptic diversity within species of trans-Andean Hyphessobrycon by leveraging species delimitation methods that do not rely on prior assignment of samples to species or lineages. Instead of delimiting species our main goal was to use these approaches to explore if there are genetic discontinuities across currently recognized species of trans-Andean Hyphessobrycon. We generated a reduced UCE dataset (36 samples; S1 Table) with no missing data (i.e., all UCEs are present in all samples) that we analyzed using two different methods: DISSECT/STACEY [62, 63] and SODA [64].

The DISSECT method [62], implemented in the package “species tree and classification estimation, Yarely” (STACEY) [63] in BEAST v.1.8.0 [65, 66], aims to identify the “minimal number of clusters of individuals” in the dataset [62, 63] as a proxy for the number of species. From our reduced UCE dataset we sub-sampled 100 UCE loci to perform the DISSECT/STACEY analysis (hereafter STACEY). We performed two independent analyses of 500,000,000 generations sampling every 50,000 generations with the following parameters CollapseHeight = 0.0001, CollapseWeight = 0.5 with a beta prior (alpha = 2, beta = 2); bdcGrowthRate = lognormal (M = 4.6, S = 2); pop-PriorScale = lognormal (M = -7, S = 2), relativeDeathRate = Beta (1,1), and collapseWeight = Beta (1,1). We evaluated convergence of the analyses using effective sampling size (ESS = > 200) for all parameters. We removed 10% of the samples as burn-in and summarized the results of STACEY in the program SpeciesDelimitationAnalyser (speciesDA.jar available at http://www.indriid.com/software.html).

We discarded 10% of the sampled trees as burn-in and summarized the posterior distribution of sampled trees into a “species or minimal clusters tree” (SMC-Tree) [63] in TreeAnnotator v.1.8.0 [66]. The probabilities of two samples belonging to the same minimal cluster of individuals were visualized in R (version 3.6.1) [67] using the function plot.simmatrix of the R scrip plot.simmatrix.R (https://github.com/scrameri/smtools/tree/master/SpeciesDelimitation) that generates a probability matrix of assignment of samples and we used the SMC-Tree as a guide to visualize the output of the plot.simmatrix script.

‘Species bOundary Delimitation using Astral’ (SODA) [64] is a multi-locus species delimitation method based on quartet frequencies implemented in ASTRAL-III. SODA tests the null hypothesis that a branch has a length of zero, and if the null hypothesis is not rejected, SODA collapses the branch subtending two samples into a “cluster species.” Branches that reject the null hypothesis (i.e., positive branch length) are then used to delimit species [64]. SODA makes two assumptions: 1) there is no gene flow and 2) no population structure; it is worth noting that if these assumptions are violated, SODA can potentially delimit species inaccurately (e.g., over split population-level structure) [64]. We generated gene trees for the reduced dataset in IQTREE-2 and we used the SMC-Tree inferred in BEAST during the STACEY analysis as a guide tree. Both gene trees and the SMC-Tree were used as input for the SODA analysis.

Biogeographic analyses

Middle America [68, 69] represents one of the most geologically complex regions on the planet, composed of island-like geological blocks (viz., Maya, Chortis, Chorotega, and Chocó, Fig 1) [69–71] that, while presently connected, have undergone multiple episodes of isolation and connectivity at different geological times [72–74]. This variation in connectivity among geological blocks likely played a role creating suitable environments at different time periods that allowed the colonization of Middle America from South America by freshwater fishes. Ostariophysan lineages (e.g., tetras and catfishes) [75] were able to colonize Middle America only after the initial closure of the Isthmus of Panama (~ 20 million years ago) [76–78].

We investigated patterns of colonization of trans-Andean Hyphessobrycon in Middle America [68, 69] using geological blocks (viz., Maya, Chortis, Chorotega, and Chocó, Fig 1) [69–71] as biogeographic units. We downloaded occurrence records of Hyphessobrycon in Middle America from the global biodiversity information facility (GBIF; www.gbif.org) [79]. We plotted these records after removing those that fell outside the known range of the genus in the region. We generated “distributional” maps for all valid species of trans-Andean Hyphessobrycon (Fig 1). The species distribution maps were generated by merging major river basins in between the two most distant occurrence records for each species in QGIS 3.4 [80]. The data for digital elevation model used was elev 30s from WorldClim [81]. River basin boundaries follow those of HydroATLAS [82, 83].

Due to the lack of fossils or geological constraints that will allow us to confidently date our phylogenomic hypothesis, we transformed our maximum likelihood tree (75% complete data matrix) into a chronogram that represents relative divergences time estimates instead of absolute ages. To generate the chronogram, we used the chronos function in the R package APE [84, 85]. Then we pruned our chronogram using the function drop.tip from the package APE [84, 85] to include only one tip per species or lineage of trans-Andean Hyphessobrycon and the species inferred as their sister South American group in our phylogenomic analyses. The pruned tree was used for downstream biogeographic analyses.

We estimated ancestral ranges in BioGeoBEARS [86] implemented in the software RASP v.4 [87]. We mapped the distribution of each species in each geological block onto the pruned tree that we used as input for RASP. Hyphessobrycon occurs exclusively in freshwater systems and there is no evidence that it can tolerate salinity. Thus, we assume that dispersal to new areas (biogeographic units) is constrained to existing freshwater connections during geologic history. River capture [88] and river anastomosis (e.g., Dias et al. [89]) are the two main processes invoked to explain dispersal (species level) or geo-dispersal (assemblage level) of freshwater taxa across adjacent basins [2]. Therefore, our biogeographic inference was constrained to only allow connections (i.e., dispersal) between adjacent biogeographic units (i.e., geological blocks; Fig 1). Based on the present-day disjunct distributions of all species of trans-Andean Hyphessobrycon (Fig 1), we constrained our analyses to allow only a maximum of two geological blocks in which one species can be distributed at a time during their evolutionary history. We compared the fit of three different parametric biogeographic models to our data based on the Akaike Information Criterion: Dispersal-Vicariance (DIVA-Like) [86, 90] Dispersal-Extinction-Cladogenesis (DEC-Like) [91] and BAYAREA-Like [92]. All the models were compared with and without the “jump-dispersal” (founder-event speciation) parameter (+J) [86].

Alignments summary

Phylogenomic inference

Different levels of missing data did not seem to have an effect on our concatenated phylogenomic inference. Overall, relationships were congruent among the three datasets (i.e., 75%, 90%, and 95% complete matrices; S1 Fig) and the majority of nodes were recovered with high support (UFBoot2 = 100, SH-aLRT = 100). Therefore, we conducted our coalescent, gene concordance factors, normalized quartet scores, and biogeographic analyses based on the gene trees and topology (concatenated and coalescent) inferred using the data matrix with a larger number of loci (75% complete data matrix). We rooted all topologies with the branch subtending to the sub-family Spintherobolinae (Figs 2 and S1) following previous phylogenomic hypotheses of the family Characidae [9].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Species tree of the family Characidae inferred with ASTRAL-III based on our 75% complete data matrix (1682 gene trees).

All nodes supported by local posterior probabilities (LPP) = 1 unless indicated in the tree. Species of Hyphessobrycon are in bold. Asterisks indicate samples from Melo et al. [9]. Illustration of male Hyphessobrycon compressus by Duangsamorn Boonwun Boyd.

https://doi.org/10.1371/journal.pone.0279924.g002

Overall, the recovered relationships were also congruent between the coalescent and concatenated inferences (Figs 2 and S1). In all analyses, we recovered the monophyly of the sub-family Stethaprioninae with strong support (LPP = 1, UFBoot2 = 100, SH-aLRT = 100; Figs 2 and S1), and recovered Hyphessobrycon as polyphyletic (n = 20; Figs 2 and S1). The analyzed species of cis-Andean Hyphessobrycon were scattered among four lineages of Stethaprioninae, some of which were more closely related to trans-Andean Hyphessobrycon than to other cis-Andean taxa. Among the cis-Andean Hyphessobrycon, we recovered the clade of ‘rosy tetras’ (LPP = 1, UFBoot2 = 100, SH-aLRT = 100) but it did not include H. compressus (Figs 2 and S1).

Coalescent and concatenation analyses recovered all species of trans-Andean Hyphessobrycon as a monophyletic group, including H. compressus (LPP = 1, UFBoot2 = 100, and SH-aLRT = 100; Figs 2 and 3 and S1). The closest relative of the trans-Andean Hyphessobrycon clade was a clade composed of Macropsobrycon xinguensis, Moenkhausia gracilima, and Hemigrammus levis. Gene concordance factors (gCF) and normalized quartet scores (NQS) analyses supported the monophyly of trans-Andean Hyphessobrycon (gCF = 63.41, NQS (T1) = 0.95; Fig 3B) whereas the alternative sets of relationships (i.e., non-monophyly of trans-Andean Hyphessobrycon) were observed at low frequencies (Fig 3B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Phylogenomic hypothesis of trans-Andean Hyphessobrycon and test of alternative sets of relationships in branches of interest.

A) Phylogenomic relationships of trans-Andean Hyphessobrycon inferred using concatenated and coalescent based methods. All nodes supported by local posterior probabilities (LPP) = 1. Support values shown from concatenation analysis, Ultrafast bootstrap (UFBoot2)/ SH-like approximate ratio test (SH-aLRT). Names of species groups and their geographic distribution in Middle America are shown. B-D) Gene concordance factors (gCF) and normalized quartet scores (NQS) for the main topology/quartet (T1) and the two alternatives quartets for a particular branch (T2 & T3): B) monophyly (T1) and non-monophyly (T2 & T3) of trans-Andean Hyphessobrycon; C) monophyly (T1) and non-monophyly (T2 & T3) of Central American Hyphessobrycon (‘panamensis’ and ‘compressus’ species groups); D) monophyly (T1) and non-monophyly (T2 & T3) of northwestern South American Hyphessobrycon (Chocó species group). EN = effective number of genes for a branch of interest, NQS = normalized quartet score, gN = number of trees decisive for the branch, gCF = gene concordance factor, gDF1 = gene discordance factor for NNI-1 branch, gDF2 = gene discordance factor for NNI-2 branch, gDFP = gene discordance factor due to polyphyly.

https://doi.org/10.1371/journal.pone.0279924.g003

Coalescent and concatenation analyses inferred three well supported groups within the trans-Andean Hyphessobrycon clade (Fig 3A). The first group included all species / lineages distributed in northwestern South America and eastern Panama (H. ecuadoriensis, H. condotensis, H. “condotensis”, and H. columbianus), hereafter referred to as the Chocó species group (LPP = 1, UFBoot2 = 100, SH-aLRT = 96.9; Fig 3A). Interestingly, samples of H. condotensis were not monophyletic (Figs 2 and 3A and S1), and individuals from the upper Atrato basin (H. “condotensis”) were more closely related to H. columbianus than to specimens collected from the Baudo and San Juan rivers (H. condotensis). The second group was composed of all samples of H. savagei, H. bussingi, and H. panamensis, hereafter the ‘panamensis’ species group (LPP = 1, UFBoot2 = 100, SH-aLRT = 100; Fig 3A). The third group included two well differentiated lineages of H. compressus (LPP = 1, UFBoot2 = 100, SH-aLRT = 100; Figs 2 and 3A and S1): the H. compressus lineage from the Usumacinta River in Guatemala and Mexico, the Yucatan Peninsula, and the Coatzacoalcos River and the H. “compressus” lineage from aquatic systems in the Izabal lake basin in Guatemala, southern Belize to northern Belize (LPP = 1, UFBoot2 = 100, SH-aLRT = 100; Figs 1–3A and S1), and H. tortuguerae, hereafter the ‘compressus’ species group (LPP = 1, UFBoot2 = 100, SH-aLRT = 100; Fig 3A). The Chocó species group was recovered as sister to a clade composed of the ‘panamensis’ and ‘compressus’ species groups: (Chocó, (‘panamensis’, ‘compressus’)). This relationship was congruently recovered by all inference methods (Figs 2 and 3 and S1) and further supported by gCF = 31.65 and NQS (T1) = 0.68 (Fig 3C). The alternative relationships among species groups: (‘compressus’, (‘panamensis’, Chocó)) or (‘panamensis’ (‘compressus’, Chocó)), were observed at relatively low frequencies when compared with the main topology, gDF1 = 4.46, NQS (T2) = 0.14, and gDF2 = 3.81, NQS (T3) = 0.18 respectively (Fig 3C).

Within the Chocó species group, H. ecuadoriensis was sister to a clade composed of H. condotensis, H. “condotensis”, and H. columbianus with high support (LPP = 1, UFBoot2 = 100, and SH-aLRT = 96.9), where H. condotensis was sister to H. “condotensis” + H. columbianus; ((H. ecuadoriensis, (H. condotensis, (H. “condotensis”, H. columbianus))) (Fig 3A). Despite the high support observed in both inference methods, the placement of H. ecuadoriensis was not unambiguously resolved. The gCF and QNS for the monophyly of the Chocó species group (T1; Fig 3A and 3D) and the alternative topologies (non-monophyly of the Chocó species group; T2 and T3, Fig 3D) were recovered in roughly similar proportions. For the monophyly of the Chocó species group (T1; Fig 3D), gCF = 16.24 and NQS = 0.40, and for the alternative topologies (i.e., non-monophyly of the Chocó species group) gCF were 14.37(gDF1) and 7.72 (gDF2), and NQS were 0.36 and 0.24 for T2 and T3, respectively (Fig 3D).

Within the ‘panamensis’ species group, both analyses of UCEs inferred H. savagei sister to H. bussingi + H. panamensis, with high support (LPP = 1, UFBoot2 = 100, SH-aLRT = 100; Fig 3A). In the ‘compressus’ species group, H. tortuguerae was sister to H. compressus + H. “compressus”, with strong support (LPP = 1, UFBoot2 = 100, SH-aLRT = 100; Fig 3A).

Investigation of cryptic diversity

The species tree inferred from the reduced dataset of 36 samples and no missing data supported the same topology of trans-Andean Hyphessobrycon species as the concatenated and coalescent trees of the complete data set (Figs 2 and S1). Both species delimitation methods suggested cryptic diversity within trans-Andean Hyphessobrycon (Fig 4). In the STACEY analysis, we investigated alternative clustering schemes after summarizing 70% of the distribution of post-burnin sampled trees. We identified three clustering schemes: the “optimal” clustering scheme recovered nine lineages within the trans-Andean Hyphessobrycon and was observed in 48% of the sampled trees, the second clustering scheme recovered 10 lineages and was observed in 15% of the sampled trees, and the third clustering scheme recovered 11 lineages and was observed in 9% of the sampled trees (Fig 4). In contrast, in the SODA analysis, we identified 15 lineages within the samples of trans-Andean Hyphessobrycon analyzed (Fig 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Investigation of cryptic diversity within species of trans-Andean Hyphessobrycon using a reduced dataset of 36 individuals and no missing data.

Cladogram on the left represents the minimum species cluster tree (MSC-Tree) inferred in the STACEY analysis. Tip labels show the catalog numbers of samples analyzed in the reduced dataset; asterisks indicate samples from Melo et al. [9] (See S1 Table). The vertical bars to the right represent the output of the species delimitation analyses–number of lineages or clusters and assignment of individuals to each one of them in SODA and STACEY. STACEY 0.48 is the “optimal” number of clusters, STACEY 0.15 is the 2^nd alternative number of clusters, and STACEY 0.09 is the 3^rd alternative number of clusters recovered. The color scheme of the bars corresponds to the color scheme in Fig 1. Illustration of female Hyphessobrycon compressus by Duangsamorn Boonwun Boyd.

https://doi.org/10.1371/journal.pone.0279924.g004

Both STACEY and SODA congruently uncovered cryptic lineages within samples of H. compressus and H. condotensis (Fig 4). It is worth noting that the optimal clustering scheme in STACEY (i.e., 0.48) grouped all samples of H. columbianus and H. “condotensis” into a single cluster, in contrast to the two minority schemes (i.e., 0.15 and 0.09) which recovered H. columbianus and H. “condotensis” as independent clusters (Fig 4). For the optimal number of clusters, the probabilities that all samples belong to their assigned “species cluster” were high and ranged from p = 0.79–1.0 (Fig 5), with the exception of H. “condotensis” and H. columbianus that showed a probability of p = 0.55 for belonging to the same cluster (Fig 5). The SODA analysis further suggested that there is cryptic diversity within H. “compressus”, H. tortuguerae, H. bussingi, H. columbianus, and H. condotensis (Fig 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Assignment probability matrix from STACEY delimitation analysis.

Each cell in the matrix shows the posterior probability of a pair of samples to belong to the same species cluster. Tip labels show the catalog number of the samples analyzed in the reduced dataset; asterisks indicate sample from Melo et al. [9] (See S1 Table). Darker colors indicate higher probability and lighter colors lower probability. The unrooted minimum species cluster tree (MSC-Tree) of trans-Andean Hyphessobrycon is shown to the left with posterior probabilities on the nodes. The posterior distribution of trees from the STACEY analysis is superimposed on the MSC-Tree. Illustration of male Hyphessobrycon compressus by Duangsamorn Boonwun Boyd.

https://doi.org/10.1371/journal.pone.0279924.g005

Biogeographic analyses

We identified the DEC-like model as the best fit biogeographical model for our dataset (Table 1). The DEC-like model was the best fit model in both model selection analyses, with and without the +J parameter (Tables 1 and S1). Following a recent critique to the usage of the +J parameter (see Ree & Sanmartín [93]) we present our results based on the analysis without the + J parameter (Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Parameter estimates of the three biogeographic models tested.

https://doi.org/10.1371/journal.pone.0279924.t001

Based on the DEC-like model, we inferred a wide ancestral range east of the Andes and in the Chocó block (AB, Fig 6) for the most recent common ancestor (MRCA) of trans-Andean Hyphessobrycon and its sister cis-Andean group (node 25 in Fig 6). We inferred six dispersal events and three vicariance events in the clade of trans-Andean Hyphessobrycon. At crown node 25, we inferred a dispersal event into the Chorotega block and a vicariant event that led to the MRCA of all trans-Andean Hyphessobrycon (Fig 6). The most likely estimated ancestral range of the MRCA of trans-Andean Hyphessobrycon included the Chocó and Chorotega blocks (node 22, Fig 6). At node 22, we inferred a vicariant event that led to the divergence of the Chocó species group and the Central American species of Hyphessobrycon. The most likely estimated ancestral ranges of the MRCA of the Chocó species group (node 21) and Central American Hyphessobrycon (node 18) were the Chocó and Chorotega blocks, respectively (Fig 6). At node 18, we inferred a dispersal event into the Chortis block and a vicariant event that led to the split of the ‘panamensis’ species group and the ‘compressus’ species group. The most likely ancestral range of the MRCA of the ‘panamensis’ group (node 17) was the Chorotega, and the Chortis block for the ‘compressus’ species group (node 15) (Fig 6). Subsequently, the MRCA of the ‘compressus’ species group dispersed into the Maya block (node 15; Fig 6). Finally, we inferred three dispersal events within the Chorotega block, one at node 16, and two at node 14 (S2 Table).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Biogeographic reconstruction of the colonization of Middle America by the genus Hyphessobrycon (i.e., trans-Andean Hyphessobrycon) based on the DEC-like model.

A) Ancestral range estimation based on the best biogeographic model DEC-like on the 75% maximum likelihood pruned topology (see Methods). Pie charts on nodes represent the probability of the estimated ancestral ranges for each clade. Boxes to the right show present-day presence/absence of trans-Andean Hyphessobrycon in the geological blocks (i.e., biogeographic units) considered in this analysis. Boxes below the phylogeny show ancestral ranges estimated that encompass two geological blocks, gray with asterisk represent other ancestral ranges. B) Map of Middle America depicting geological blocks and their letter codes. Numbers on circles indicate nodes in A for which we inferred a dispersal event (colonization) to a new geological block. C) Schematic representation of major biogeographic events during the colonization of Middle America by Hyphessobrycon. Circles represent nodes on the phylogeny of trans-Andean Hyphessobrycon, node numbers above or below circles, and their most likely ancestral range estimated (see A), circles without node number represent an inferred event (i.e., range expansion) that occurred in the previous node.

https://doi.org/10.1371/journal.pone.0279924.g006

Phylogenetic relationships

Concatenated and coalescent analyses of 1682 UCE loci are concordant with previous hypotheses that species currently recognized as Hyphessobrycon are polyphyletic (Fig 2; e.g., [14, 17, 31]). We recovered strong support for the monophyly of the trans-Andean species of Hyphessobrycon (Figs 2 and 3 and S1). A clade comprised of Macropsobrycon xinguensis, Hemigrammus levis, and Moenkhausia gracilima all distributed in the Amazon River basin, was inferred as sister to the trans-Andean Hyphessobrycon clade (Figs 2 and 3). Despite the strong support for this sister relationship with trans-Andean Hyphessobrycon (LPP = 1, UFBoot2 = 100, SH-aLRT = 100; Figs 2 and S1), an alternative hypothesis (i.e., H. diancistrus + trans-Andean Hyphessobrycon) was observed in roughly equal frequency (S2 Fig). Recent phylogenomic work of tropical characoids that included a few representatives of Hyphessobrycon recovered a clade of H. compressus + H. diancistrus sister to M. xinguensis, H. levis, and M. gracilima [9]. Furthermore, a close relationship of H. compressus to other cis-Andean Hyphessobrycon species (i.e., H. bayleyi, H. diancistrus, and H. otrynus) from the Amazon basin has been suggested based on unpublished molecular data, see Lima et al. [94]. Our phylogenomic hypothesis unambiguously placed H. compressus in a clade exclusively composed of trans-Andean Hyphessobrycon (Figs 2 and 3 and S1) and challenges previous hypotheses that suggest that H. compressus is closely allied with cis-Andean taxa [14, 31, 32, 94]. Undoubtedly, future work that includes more taxa of the subfamily Stethaprioninae will enable us to more confidently place the trans-Andean Hyphessobrycon within the most diverse subfamily of Characidae [21].

Phylogenetic relationships and cryptic diversity of trans-Andean Hyphessobrycon

Within the trans-Andean Hyphessobrycon, our results support the recognition of three subclades: the Chocó species group, the ‘panamensis’ species group, and the ‘compressus’ species group. Our analysis recovered the Chocó species group sister to a clade composed of the ‘panamensis’ + ‘compressus’ species groups (Fig 3A). Furthermore, our species delimitation approaches uncovered cryptic diversity in two species of the Chocó and the ‘compressus’ species groups: H. condotensis and H. compressus, respectively (Figs 4 and 5). Furthermore, we interpret the inferred cryptic diversity within H. “compressus”, H. tortuguerae, H. bussingi, and H. columbianus by the SODA analysis (Fig 4) as population-level structure (see Methods). The Chocó species group, which is restricted to aquatic systems in the Chocó block in northwestern South America and eastern Panama (Fig 1) [70, 71], is composed of H. ecuadoriensis, H. condotensis, H. columbianus, and a lineage named H. “condotensis” (Figs 4 and 5). It is worth noting that H. “condotensis” is closely related to H. columbianus, and both lineages are distributed in aquatic systems that drain to the Atlantic slope of northwestern South America and eastern Panama. In contrast, samples assigned to H. condotensis were collected in the San Juan and Baudo rivers which drain into the Pacific slope of Colombia (Fig 1). Despite the lack of samples of H. daguae in our study, we hypothesize that this species belongs to the Chocó species group because of the observed correlation of phylogeny (i.e., species groups) with geographic distributions (i.e., geological blocks). Furthermore, close relationships of H. daguae with species of Hyphessobrycon in northwestern South America has been proposed based on morphological characters, see Ota et al. [26]. The taxonomy of Hyphessobrycon species distributed in the Chocó block has remained in flux [26, 33, 95] and the presence of cryptic diversity has been suggested [96]. Our work uncovered cryptic diversity in the region (Figs 4 and 5) and disagrees with the current taxonomic hypothesis of four valid species of Hyphessobrycon (i.e., H. columbianus, H. condotensis, H. daguae, and H. ecuadoriensis) distributed in the Chocó block (see Ota et al. [26]). For example, the species H. sebastiani was proposed as a junior synonym of H. condotensis based on morphological and coloration characters [26]. Our robust phylogenomic hypothesis (Figs 2 and 3 and S1) coupled with the investigation of cryptic diversity (Figs 4 and 5) do not support the recognition of H. condotensis as a single widespread species in the Pacific and Atlantic slope of Colombia (Fig 1) proposed by Ota et al. [26]. We propose two hypotheses for the identity of the cryptic H. “condotensis” lineage: a) it belongs to the previously recognized H. sebastiani, which has been reported to be distributed in the Upper Atrato basin [33] or b) it represents a newly recovered lineage of the Chocó species group. Further work is needed to clarify the taxonomic identity of H. “condotensis” in the Upper Atrato River.

Finally, despite the fact that our coalescent and concatenation analyses both recover the monophyly of the Chocó species group (Figs 2 and S1), gCF and NQS suggest that the placement of H. ecuadoriensis is not unambiguously resolved. Specifically, we observed a similar proportion of gene trees that support recognition of the Chocó species group excluding H. ecuadoriensis. This hypothesis resolved H. ecuadoriensis as the sister to all trans-Andean Hyphessobrycon (T2; Fig 3D). Future work that includes more robust sampling is needed to better understand and recognize the diversity of Hyphessobrycon in the Chocó block and will help to clarify the phylogenetic relationships among this group and within trans-Andean Hyphessobrycon.

The ‘panamensis’ species group is comprised of three species, H. savagei, H. panamensis, and the recently described H. bussingi [26]. These three species are mainly distributed in the Pacific (H. savagei) and Atlantic (H. bussingi and H. panamensis) slopes of the Chorotega block [71]. The sister relationship of the two Atlantic species relative to the Pacific one suggests a possible vicariant role of the orogenic process giving rise to the Talamanca mountain range [97] that split the MRCA of the ‘panamensis’ species group into two Pacific and Atlantic lineages. Following divergence, speciation in situ may have occurred within the Atlantic slope of the Chorotega block. The ‘compressus’ species group is comprised of three lineages: H. compressus, H. “compressus”, and H. tortuguerae. The recovered close relationships between these three lineages support the hypothesis of Böhlke [30], who allies H. compressus, H. tortuguerae, and H. milleri (currently synonymized under H. compressus, see below) regardless of differences in coloration among other characters [30]. The uncovering of cryptic diversity within H. compressus (Fig 1) is remarkable considering a recent and thorough systematic work that redescribed H. compressus and recognized it as a widespread species (Fig 1) across Mexico, Guatemala and Belize “without detected variations in meristic and morphometric data” [31]. These results led to the recognition of H. milleri as junior synonym of H. compressus [31]. Interestingly, our sampling across the distribution of H. compressus (Fig 1) mirrors the distribution of specimens analyzed in the previous morphological study (see fig. 7 of Carvalho & Malabarba [31]), which highlights the conserved morphology of some lineages of trans-Andean Hyphessobrycon (Figs 4 and 5). We propose two plausible hypotheses for the identity of the cryptic H. “compressus” lineage a) it belongs to the previously recognized H. milleri, hence extending its distribution into southern and northern Belize, or b) it represents a newly uncovered lineage of the ‘compressus’ species group. Specimens of Hyphessobrycon in the Motagua River basin, where the type locality of H. milleri is located, have not been collected since 1974 despite recent collecting efforts by the authors (DJE and CDM). The inclusion of samples from the Motagua basin for molecular analyses would be key for testing our proposed hypotheses regarding the identity of the H. “compressus” lineage.

Finally, the relationships recovered among the three species groups, (Chocó (‘panamensis’, ‘compressus’)), do not support the recent hypothesis of relationships for some species of trans-Andean Hyphessobrycon that proposed the ‘H. panamensis species-group’ sensu Ota et al. [26] is comprised of H. bussingi, H. columbianus, H. condotensis, H. daguae, H. panamensis, and H. savagei. Coloration patterns alone or in combination with morphological characters have been used to propose relationships among species of trans-Andean Hyphessobrycon. For example, it has been suggested that pigmentation (or lack thereof) on the dorsal fin reflects relationships among species of trans-Andean Hyphessobrycon including the ‘H. panamensis species-group’ [26, 31, 32]. However, our robust phylogenomic hypotheses do not support the monophyly of the ‘H. panamensis species-group’ sensu Ota et al. [26] and cast doubt on the use of these coloration patterns as phylogenetically informative characters for trans-Andean Hyphessobrycon. Finally, it is worth noting that for the Central American species of Hyphessobrycon (i.e. H. compressus, H. tortuguerae, H. savagei, H. milleri, and H. panamensis), three morphological synapomorphies that unite these species have been proposed: a) premaxilla with seven teeth in the inner row, b) round foramen in the ventral region of the quadrate, and c) two foramina in the ventral margin of the epiotic bone, see García-Alzate et al. [33]. Future work is necessary to test if these synapomorphies are diagnostic of Hyphessobrycon sensu stricto.

Colonization of Middle America

The best-fit biogeographical model first inferred a vicariant event separating sister cis-Andean taxa and trans-Andean Hyphessobrycon (node 25; Fig 6). This event was plausibly promoted by orogenic processes in northwestern South America, like the uplift of the Andes during the Miocene [98, 99], that played an important role shaping the riverscape that promoted the origin of trans-Andean ichthyofauna [78, 100–102].

Our results support a single colonization event of Middle America by Hyphessobrycon. The use of geological blocks as biogeographic units allowed us to uncover a stepwise pattern of colonization from South (Chocó block) to North (Maya block), followed by in-situ diversification within each geological block (Fig 6). Our ancestral range estimation inferred that the MRCA of all trans-Andean Hyphessobrycon was distributed in an area comprising the Chocó and Chorotega blocks (node 22; Fig 6). In this region, another vicariant event led to the separation of the Chocó and ‘panamensis’ species groups. Our biogeographical inferences together with the present-day distribution of the Chocó and ‘panamensis’ species groups agree with the early colonization model of ostariophysan fishes of Bermingham and Martin [103]. This model suggests an early colonization into Central America (Chorotega block) during the late Miocene, followed by extinction (in some taxa) in some river basins due to changes in sea levels during the Pliocene [77, 78, 100, 102–105]. Furthermore, our current understanding of the complex landscape in lower Middle America and the timing of the initial closure of the Isthmus of Panama (~ 20 million years ago) [76, 106] provides evidence of connectivity between the Chocó and Chorotega blocks during the Miocene.

From the Chorotega block, we inferred a range expansion into the Chortis block followed by a vicariant event that separated the ‘panamensis’ and ‘compressus’ species groups. The range fragmentation of the MRCA of the Central American species of Hyphessobrycon may be explained by a marine corridor across the Nicaragua depression, which was located between the Chorotega and Chortis blocks until the Pliocene [72]. Global sea level changes [107] have impacted the landscape of Middle America (see Bagley and Johnson [108] and references therein). Due to physiological constraints of Hyphessobrycon (i.e., salinity intolerance), a marine corridor likely extirpated ancestral populations and limited gene flow between populations on either side of the barrier that gave origin to the ‘panamensis’ and ‘compressus’ groups. Finally, present-day disjunct distributions between species in the ‘compressus’ group (Fig 1) preclude us from making inferences regarding the colonization route from the Chortis into the Maya block.