Specimens were collected under a permanent permission number 13843-1 from MMA/IBAMA/SISBIO and subsequently preserved in 95% ethanol. We included 146 specimens spanning all 13 species of Prochilodus collected across all South America plus Semaprochilodus taeniurus to root the trees (total 147 taxa). We sequenced barcodes for 19 specimens and supplemented the matrix with 127 additional barcodes of Prochilodus available at the public genetic databases Genbank (www.ncbi.nlm.nih.gov/) and Barcode of Life Database (BOLD; www.boldsystems.org/). Supplementary Table S1 contains voucher and locality information and accession numbers for databases.

Genomic DNA was extracted from muscle tissues preserved in 95% ethanol with a DNeasy Tissue kit (Qiagen Inc.; http://www.qiagen.com) according to the manufacturer's instructions. We obtained partial sequences of the mitochondrial gene cytochrome oxidase c subunit I by amplifying via polymerase chain reaction (PCR) using the primer described in the literature (Melo et al., 2011) and modifying reaction steps as follow: 12.5 μl as a total volume with 9.075 μl of double-distilled water, 1.25 μl 5x buffer, 0.375 μl MgCl₂ (50 mM), 0.25 μl dNTP mix, 0.25 μl of each primer at 10 μM, 0.05 μl Platinum Taq DNA polymerase enzyme (5 units/μl, Invitrogen; www.invitrogen.com) and 1.0 μl genomic DNA (10–50 ng). The PCR consisted of an initial denaturation (4 min at 95°C) followed by 28–30 cycles of chain denaturation (30 s at 95°C), primer hybridization (30–60 s at 52–54°C), and nucleotide extension (30–60 s at 72°C). After the visualization of the fragments using 1% agarose gel, we performed the sequencing reaction using dye terminators (BigDye™ Terminator v 3.1 Cycle Sequencing Ready Reaction Kit, Applied Biosystems; http://www.appliedbiosystems.com) purified again through ethanol precipitation. We then loaded the samples onto an automatic sequencer ABI 3130-Genetic Analyzer (Applied Biosystems) at the São Paulo State University, Brazil.

We assembled and edited the newly generated consensus sequences in Geneious 7.1.9 (Kearse et al., 2012) and aligned the whole matrix with Muscle (Edgar, 2004). This matrix contains 147 taxa (146 Prochilodus plus one Semaprochilodus) and 648 bp. To evaluate the occurrence of substitution saturation, the index of substitution saturation in asymmetrical (Iss.cAsym) and symmetrical (Iss.cSym) topologies were estimated in Dambe 5.3.38 (Xia, 2013). We used PartitionFinder 1.1.0 (Lanfear et al., 2012) to select the best-fit model of nucleotide evolution for our dataset.

Species were previously identified following the most recent and complete taxonomic revision (Castro and Vari, 2004), and lineages were proposed based on subsequent topologies. Most available sequences are from vouchers already identified by the first author (e.g., Melo et al., 2016b) or from previous studies with endemic species (Carvalho et al., 2011; Rosso et al., 2012; Pereira et al., 2013; Díaz et al., 2016). We then generated overall and pairwise values of genetic distance based on Kimura-2-parameters (K2P)+Gamma using Mega 7.0 (Tamura et al., 2013) and a neighbor-joining tree (NJ) with 1,000 replicates of bootstraps using Geneious 7.1.9. We also performed a maximum likelihood (ML) analysis under RAxML HPC-PTHREADS-SSE3 (Stamatakis, 2006) using five random parsimony trees with the GTRGAMMA model (Stamatakis et al., 2008) without rooting and with other parameters at default. We used the autoMRE function to generate pseudoreplicates through MRE-based stopping criteria (Pattengale et al., 2009) that ran a total of 650 replicates. Stopping criteria determine when enough replicates have been generated so that robust bootstraps under ML analysis become computationally practical (Pattengale et al., 2009).

An ultrametric gene tree was generated in a Bayesian inference with Beast 1.8.0 (Drummond et al., 2012) using two independent runs of 50 millions generations sampling trees every 5000th generation. Convergence was indicated by Tracer v1.5 (Rambaut et al., 2014) with estimated sample sizes (ESS) superior to 200. An appropriate number of trees (first 10%) from each run was discarded as burn-in and the MCMC samples was generated using the maximum clade credibility (MCC) topology in TreeAnnotator v1.4.7 (Drummond et al., 2012) and visualized in FigTree v1.4.3.

The general mixed Yule coalescent (GMYC) method (Pons et al., 2006; Fujisawa and Barraclough, 2013) was performed using the ultrametric gene tree estimated with the exponential growth coalescent model (Griffiths and Tavaré, 1994) and the lognormal relaxed clock model (Drummond et al., 2006), which assumes that the rates of molecular evolution are uncorrelated but log-normally distributed among lineages. Species delimitation through GMYC model was conducted using standard parameters [interval = c(0, 10)] and a single threshold that specifies the transition time between to within species branching. Such analysis was conducted with the package splits (Species Limits by Threshold Statistics; http://r-forge.r-project.org/projects/splits) in R v.3.0.0 (R Development Core Team, 2013). GMYC appears to be useful for single-locus analysis (Fujisawa and Barraclough, 2013) but depends on the availability of additional data/analyses from independent characters (Esselstyn et al., 2012). Additionally, we used the Bayesian Poisson Tree Processes model (bPTP) (Zhang et al., 2013) in the bPTP webserver (http://species.h-its.org/ptp/) under default parameters. bPTP does not require an ultrametric gene tree and uses, instead, a nexus tree as input file with branch lengths representing the number of nucleotide substitutions (Zhang et al., 2013). We used a nexus MCC tree generated in Beast 1.8.0 (Drummond et al., 2012) as input file and ran 500,000 generations (thinning = 500). We also used a clustering species delimitation analysis through the Automatic Barcode Gap Discovery (ABGD; Puillandre et al., 2012) that automatically defines sequences into hypothetical candidate species based on confidence limits for intraspecific divergence. We used a pairwise distance matrix generated in Mega 7.0 (Kumar et al., 2016) through K2P+G model and 1,000 pseudoreps as input file into the ABGD webserver (wwwabi.snv.jussieu.fr/public/abgd/abgdweb.html) with other parameters left at defaut.

Population genetic analyses were conducted in order to detect levels of genetic variance among haplotypes. We excluded four taxa and excized flanking regions with elevated missing data to properly run those analyses. This reduced matrix contained 143 taxa and 465 bp. Each mitochondrial lineage previously determined by distance and likelihood analyses was treated as a distinct population. We used DnaSP v.5.10.01 (Librado and Rozas, 2009) to obtain the number of polymorphic sites, haplotype number, and nucleotide/haplotype diversity. In Arlequin 3.5.1 (Excoffier and Lischer, 2010), each mitochondrial lineage was set as a single population with the following hypothetic group structuring (group 1 = outgroup; group 2 = lineage 1; group3 = lineage 2; group 4 = lineages 3, 4, and 5; group 5 = lineages 6, 7, and 8) based on the arrangement from ML and Bayesian trees. We ran an analysis of molecular variance (AMOVA; Excoffier et al., 1992) with 1,000 permutations using conventional F-statistics and generated the haplotype network using the median joining analysis (Bandelt et al., 1999) incorporated in PopART 1.7 (Leigh and Bryant, 2015).

Results from the species delimitation analysis revealed the presence of eight genetic lineages covering 13 valid species of Prochilodus, in which four lineages (1, 3, 4, and 5) are structured by only one species and the other four lineages (2, 6, 7, and 8) include more than one species. Topologies (Figure 1, Supplementary Figures S1, S2) are quite similar to the molecular phylogeny of Prochilodontidae (Melo et al., 2016a), likely due the locus selection. Prochilodus vimboides (lineage 1), for example, splits from the most recent common ancestor of all other Prochilodus, although the biogeographic implications for this result still requires a more detailed, time-calibrated analysis of the Prochilodontidae. Other example is the structuring of P. harttii (lineage 4) and P. argenteus (lineage 5), evidencing distinct genetic lineages even with a recent evidence of hybridization (Sales et al., 2018). In contrast with the molecular phylogeny, results indicate that P. mariae is an exclusive cluster, suggesting inconsistencies in the phylogenetic placement of the species (Castro and Vari, 2004; Melo et al., 2016a). A phylogeographic study found that P. mariae diverged from P. cf. rubrotaeniatus in a very recent cladogenesis (Turner et al., 2004), which does not match the Orinoco-Amazon vicariant event resulted from the rise of Vaupes Arch during the Late Miocene (Lujan and Armbruster, 2011).

Our findings suggest the recognition of only one trans-Andean species, in which Prochilodus magdalenae from Río Magdalena remains nested within P. reticulatus from Lago Maracaibo (lineage 2) as proposed by the molecular phylogeny (Melo et al., 2016a). Species limits among them involve subtle differences in the range and modal values of number of lateral line scales, number of predorsal scales, and number of vertebrae (Castro and Vari, 2004). A further phylogeographic study involving samples from Atrato, Cauca-Magdalena, and Maracaibo might help to elucidate the allopatric distribution of this mitochondrial lineage.

The Amazon basin harbors two distinct mitochondrial lineages of Prochilodus nigricans (lineages 6 and 7). A recent study detected population structure in western populations of P. nigricans (Madeira and Purus) compared to those from mainstream Rio Amazonas (Machado et al., 2017) despite the lack of samples from eastern tributaries. Interestingly, there is ecological evidence of two distinct migration patterns of P. nigricans in the Amazon basin (Araújo-Lima and Ruffino, 2003) that might explain our results. The first involves lateral migrations from floodplain lakes to the mainstream Rio Amazonas with subsequent migration upstream to breeding and spawning (Fernandes, 1997), and the second involves only upstream migrations to upper Rio Tocantins or Araguaia to spawning and downstream migrations to feeding (Carvalho and Mérona, 1986), the latter similar to the well-known pattern observed for P. argenteus (Godinho and Kynard, 2006) and P. lineatus (Agostinho et al., 2004).

Our results do not support the presence of multiple species within lineage six. Prochilodus britskii, a morphologically distinct species, appears for the first time embedded within the lineage, differently from the position as sister to P. mariae (Melo et al., 2016a). Both P. brevis and P. lacustris from northeastern Brazil are distinguished from the species pair P. nigricans and P. rubrotaeniatus by radial subdivision patterns on body scales (Castro and Vari, 2004). Morphological features diagnosing each of them include overlapped counts of lateral line scales, number of horizontal scale rows below lateral line, and number of circumpeduncular scale rows (Castro and Vari, 2004).

Castro and Vari (2004) redescribed P. nigricans by examining almost one thousand Amazonian specimens including type specimens. They designated a neotype from Lago Janauacá at the right margin of Rio Solimões near Manaus in Brazil. Twenty-one individuals of P. nigricans from Manaus appear within lineage seven along with specimens from western Amazon. Therefore, this cluster likely constitutes the genetic lineage of the neotype. The position of P. rubrotaeniatus still represents a lacuna in our knowledge due the presence of the species in both lineages six and seven. Based on a previous phylogeographic study (Turner et al., 2004), Albert et al. (2011) suggest that P. rubrotaeniatus represents an example of paraspecies that gave rise to the endemic P. mariae. Paraspecies are paraphyletic, geographically widespread species that originates another peripheral isolated species without becoming extinct (Ackery and Vane-Wright, 1984; Albert et al., 2011). The phylogenetic evidence (Melo et al., 2016a) and the results arrived herein refute such hypothesis and instead, indicate that the present concept of P. rubrotaeniatus constitute more than one genetic lineage.

In the Brazilian Shield, Prochilodus costatus share the same mitochondrial cluster with P. lineatus (lineage 8), which again corroborates the molecular phylogeny (Melo et al., 2016a) and a mitogenome analysis (Chagas et al., 2015). Analyzed specimens of P. costatus from Rio Pandeiros/São Francisco are remarkably distant (~2,500 linear km) from analyzed specimens of P. lineatus from Rosario in Argentina. These results agree with previous population genetic and phylogeographic studies that show high genetic diversity and low population divergence (Sivasundar et al., 2001; Carvalho-Costa et al., 2008; Melo et al., 2013; Ferreira et al., 2017). A phylogeographic study of P. lineatus, for example, found strong similarity among mitochondrial control regions between samples from the lower Rio Paraná in Argentina and upper Rio Paraná in Brazil (Sivasundar et al., 2001). Overlapped counts of lateral line scales, number of vertebrae and allopatry slightly discriminate the two species (Castro and Vari, 2004), which is clearly not supported herein. In addition, our evidence indicates that future population genetic studies of one or another species should include members of both nominal species.

Results indicate very little mitochondrial divergence among lineages of Prochilodus and provide evidence that distantly sampled specimens, in various instances, correspond to a single mitochondrial lineage. The most plausible hypothesis that might explain such result is that migration affects species diversification. Indeed, migration has been used to explain high levels of gene flow and low population structure in Prochilodus (e.g., Sivasundar et al., 2001; Carvalho-Costa et al., 2008; Melo et al., 2013; Ferreira et al., 2017). This is supported by ecological data from fish tagging that found migratory routes of >120 km for P. argenteus along the Rio São Francisco (Godinho and Kynard, 2006) and 250 km of Prochilodus sp. in that same basin (Paiva and Bastos, 1981).

Migration and gene flow directly influence morphological stasis (Stanley, 1979). This raises some questions about how migration patterns have influenced population diversification without morphological change in Prochilodus. Would distinct environmental settings be responsible for distinct movement behaviors along their evolutionary history? López-Fernández and Albert (2011) suggest that massive prochilodontid migrations evolved during the Oligocene, before the separation of the paleo-Amazon-Orinoco river basin, and that posterior vicariant events allowed their successful colonization throughout major Neotropical basins. The two allopatric lineages of P. nigricans and their two migration patterns (Carvalho and Mérona, 1986; Fernandes, 1997) support López-Fernández and Albert (2011)'s conclusion and reinforce the fact that a lineage (or population) once fragmented tends to search for ecological adaptation in distinct environmental conditions. Colonization of Neotropical habitats in upland rivers requires adaptation to a strong selective pressure by acquiring specific morphological innovations (Silva et al., 2016b) or behavior specializations, which appears to be the case of Prochilodus.

It is noteworthy, however, that migration is not the exclusive factor aging disfavoring species diversification in Prochilodus. There are, at least, two more plausible explanations for the observed low genetic variation. Hybridization between the native P. harttii and the introduced P. argenteus has been genetically identified in the Rio Jequitinhonha recently (Sales et al., 2018). Although hybridization between native species has not been documented yet, the process might be included as another hypothesis to explain our results, at least for sympatric species. Another plausible hypothesis would be recent episodes of species diversification that did not allow accumulation of haplotype variation. Despite a molecular phylogeny is available (Melo et al., 2016a), the lack of a time-calibrated tree does not allow us to support this hypothesis with better confidence. Testing those three most plausible hypotheses to explain the little divergence among lineages of Prochilodus is thus a matter of further research.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.