Astyanax Baird & Girard, 1854 is one of the most diversified and taxonomically complex genus within the Characidae family (Characiformes), including 125 valid species widespread throughout nearly the entire Neotropical region (Rossini et al., 2016; de Pinna et al., 2018; Fricke et al., 2022). This genus comprises small species of about 40 to 200 mm standard length (Garutti, 1998), occurring in a wide diversity of niches and aquatic environments within freshwater drainages from the southern United States to central Argentina (Eigenmann, 1921; Bertaco, Garutti, 2007).

Characterized by presenting high phenotypic plasticity and adaptation to distinct environmental conditions (Orsi et al., 2004), Astyanax species are among the most important components of the freshwater food web, with significant participation in the diet of large predator fishes (Prioli et al., 2002), being usually dominant in headwaters and small tributaries (Bertaco, Lucena, 2010). Within the Brazilian Shield, the genus is commonly found in large river systems (e.g., Amazon, La Plata, and São Francisco) and in the northeastern Brazilian coastal basins, including the Paraguaçu River basin.

The Paraguaçu River basin is considered one of the largest basins in northeastern Brazil (Higuchi et al., 1990) and an extremely relevant drainage of the Northeastern Mata Atlântica freshwater ecoregion (NMAF, ecoregion 328, sensu Abell et al., 2008; Camelier, Zanata, 2014a). The fish fauna of the basin has been recognized by its high level of endemism (Buckup, 2011; Camelier, Zanata, 2014a; de Pinna et al., 2018). Recently, new species have been described for this basin, including different fish groups, such as Astyanax (e.g., Camelier, Zanata, 2014b; Zanata et al., 2017, 2018; Burger et al., 2019), Characidium Reinhardt, 1867 (e.g., Zanata, Camelier, 2015; Melo, Espíndola, 2016), Copionodon de Pinna, 1992 (e.g., de Pinna et al., 2018) Moenkhausia Eigenmann, 1903 (e.g., Benine et al., 2009), and Rhamdiopsis Haseman, 1911 (e.g., Bockmann, Castro, 2010).

More than ten species of Astyanax are currently reported as occurring in the Paraguaçu hydrographic system (Santos, Caramaschi, 2007, 2011). From this total, six species are endemic to the upper Paraguaçu course and have allopatric distribution, occurring in different tributaries: A. brucutu Zanata, Lima, Dario & Garhard, 2017, Pratinha River (Zanata et al., 2017); A. epiagos Zanata & Camelier, 2008, Jacuípe River (Zanata, Camelier, 2008); A. hamatilis Camelier & Zanata, 2014, Utinga, Una, and São José rivers (Camelier, Zanata, 2014b); A. lorien Zanata, Burger & Camelier, 2018, Santo Antônio River; A. rupestris Zanata, Burger & Camelier, 2018, Coisa Boa and Cumbuca rivers (Zanata et al., 2018); and A. sincora Burger, Carvalho & Zanata, 2019, Tremedal stream (Burger et al., 2019). Furthermore, according to Zanata et al., (2018), the Piabinha River shelters a morphotype tentatively identified by the authors as Astyanax aff. rupestris, due to divergences in some morphological characters when compared to A. rupestris. The high richness within Astyanax and the fact of being traditionally defined by a combination of non-exclusive characters (see Eigenmann, 1921), added to its recognized phenotype plasticity (Orsi et al., 2004), occasionally hinders accurate species identification. Consequently, some taxa are frequently identified only at the generic level or into species complexes (e.g., Moreira-Filho, Bertollo, 1991; Garutti, Britski, 2000). A recent integrative phylogeny (Terán et al., 2020) recovered species attributed to Astyanax in different subfamilies and genera, including the resurrected Psalidodon Eigenmann, 1911 and a new genus, Andromakhe Terán, Benitez & Mirande, 2020 (Terán et al., 2020). Dagosta, Marinho, (2022) argue that although this study has been efficient in recovering the polyphyletic nature of Astyanax, it failed in providing consistent diagnosis characters for the proposed clades. None of the species evaluated here were analyzed by Terán et al., (2020), except A. brucutu that, due to the lack of molecular data, was inserted as incertae sedis in Gymnocharacini. In view of that, the species is herein assigned to Astyanax.

It is well known that, given the remarkable richness and phenotypic plasticity observed in the Neotropical freshwater ichthyofauna (Wimberger, 1992; Reis et al., 2016), and its high number of cryptic species (Piggott et al., 2011), the genetic analysis is a powerful tool for improving our knowledge on taxonomy and evolution of this group (Bellafronte et al., 2013; Costa-Silva et al., 2015; Anjos et al., 2020). Different DNA-based approaches, such as DNA barcode (Hebert et al., 2003; Ward, 2009), molecular species delimitation (Pons et al., 2006; Puillandre et al., 2012; Ratnasingham, Hebert, 2013), and molecular phylogeny analyses (Edwards, 2009), have been successfully used for defining Molecular Operational Taxonomic Units (MOTUs) and characterizing hidden biodiversity within Neotropical freshwater fish (e.g., Ramirez, Galetti Jr., 2015; Carvalho et al., 2011; Pereira et al., 2011, 2013; Machado et al., 2016; Ramirez et al., 2017; Silva-Santos et al., 2018; Souza et al., 2018; Lopes et al., 2020).

Here, we performed species delimitation analyses in four recently described species of Astyanax plus the morphotype A. aff. rupestris, all endemic to the upper Paraguaçu River basin. We aimed to produce a DNA barcode reference library for the focal species and to investigate the existence of hidden diversity, contributing thus to a better knowledge of this relevant fish group and its diversification. Using mitochondrial and nuclear sequences, we combined single and multilocus-based methods to carry out genetic analyses. Our sequence data were compared to those that had already been published in Astyanax species studies.

Biological sampling. Biological samples of four endemic species and one morphotype of Astyanax were collected from 76 specimens distributed along six tributaries in the upper Paraguaçu River basin, Bahia, Brazil (Fig. 1). The material analyzed included A. brucutu from the Pratinha River, Iraquara (n = 3); A. epiagos from the Ferro Doido River, Jacobina (n = 5); A. lorien from the Preto River, Palmeiras (n = 15); A. rupestris from the Coisa Boa River, Andaraí (n = 17) and Cumbuca River, Mucugê (n = 3); and the morphotype Astyanax aff. rupestris from the Piabinha River, Mucugê (n = 33). Fin fragments were sampled from each specimen, using tweezers and scissors, and then stored in ethanol (95%) in a freezer at 4^oC. Vouchers were deposited in the ichthyological collection of the Museu de História Natural da Bahia, Salvador, Bahia, Brazil. All information related to the sampling localities, specimens, and vouchers is available in Tab. S1.

We complemented our biological sampling by downloading 1,792 COI sequences available in the BOLD system database (http://www.boldsystems.org/, accessed on March 31, 2020) for 66 nominal Astyanax species of several hydrographic basins from localities informed by the database depositors (Tab. S2). That dataset included A. hamatilis from São José River (n = 7); and unidentified specimens of Astyanax sp. from Coité (n = 2) and Piabinha (n = 2) rivers from the upper Paraguaçu basin. Altogether we analyzed five from the six Astyanax species endemic to the upper Paraguaçu River, except A. sincora, that was not collected and was not available in the BOLD system as well.

FIGURE 1 |

DNA isolation, amplification, purification, and sequencing. DNA extraction was performed using buffer saline protocol (Aljanabi, 1997), and DNA was quantified using a Biophotometer (Eppendorf, Hamburg, Germany). Partial Cytochrome c Oxidase subunit I (COI), Cytochrome b (Cytb) and the first intron of the S7 ribosomal protein (S7) genes were amplified using the following oligonucleotides: COI FishF1 and COI FishR1 (Ward et al., 2005), AnosCytBF and AnosCytBR (Ramirez, Galetti, 2015), and S7RPEX1F and S7RPEX2R (Chow, Hazama, 1998). Polymerase Chain Reactions (PCRs) were performed according to their respective authors.

The amplified products were checked on agarose gel 1% by electrophoresis, and then purified with a polyethyleneglycol (PEG) 20% protocol (Lis, 1980). Sequencing was run on an automated sequencer ABI3730XL (Applied Biosystems, Little Chalfont, UK), and all sequences were aligned and edited with the Geneious 6.1.6c software (Kearse et al., 2012). The plugin “find heterozygotes” of this software was used with a 0.80 threshold in order to identify heterozygous positions and assign ambiguity codes in the nuclear sequences, such as eventual NUMTS (nuclear mitochondrial DNA segments). The S7 haplotypes were estimated using the SEQPHASE web tool (Flot, 2010). COI sequences were deposited in the BOLD system under Project name ASTBA. Cytb and S7 sequences were deposited in the GenBank database (https://www.ncbi.nlm.nih.gov/) under specific accession numbers as shown in the Tab. S1.

Single locus analyses. To obtain a general picture of genetic relationships within Astyanax, we first implemented a broad Bayesian inference analysis with BEAST 2.4.6 (Heled, Drummond, 2010) using a large COI sequence dataset, representing the four species studied herein and other 66 nominal Astyanax species obtained from BOLD. Two independent runs were performed following the parameters: 100 million generations (Markov chain Monte Carlo, MCMC), sampling every 10,000, a strict lognormal clock for all partitions, and the Yule speciation model. The best-fitting model (GTR+I+G) was selected under the Bayesian Information Criterion (BIC) by jModeltest 2 (Darriba et al., 2012). A consensus tree was combined and resampled in LogCombiner with 30% burn-in, and then summarized in TreeAnnotator using BEAST 2.4.6 (Heled, Drummond, 2010). An effective sample size (ESS) of 200 or higher was required for all parameters and checked in TRACER 1.6 (Rambaut et al., 2014).

Based on the COI tree, only the species recovered in a single clade, which included the four targeted Astyanax species, were hereafter analyzed. For this new dataset, we implemented three species delimitation approaches using the COI sequences: Barcode Index Number (BIN, Ratnasingham, Hebert, 2013), Automatic Barcode Gap Discovery (ABGD, Puillandre et al., 2012), and General Mixed Yule Coalescent (GMYC, Pons et al., 2006). The BIN analysis was performed automatically in the BOLD system. Our sampling was assembled in a preexisting BIN database or assigned to a new BIN (Ratnasingham, Hebert, 2013). For the ABGD we used the K2P (Kimura-2-parameters) modified parameters (Pmin = 0.04, Pmax = 0.1, relative value gap X = 0.1), and 100 steps. The GMYC analysis was implemented in the SPLITS package for R statistical software (R Development Core Team, 2017), using a single threshold under the standard parameters (interval = c(1,10)). This analysis uses an ultrametric tree to establish species limits based on the Yule (pure-birth) and Kingman models (coalescence), and to calculate the probability of splits in a lineage based on speciation rates (Ratnasingham, Hebert, 2013). As input, we used an ultrametric tree obtained with a lognormal relaxed clock, birth-death speciation model, HKY + G substitution model, 50 million MCMC sampling every 5,000 and burn-in of 10% in BEAST 2.4.6. Convergence was assessed by estimating the effective sampling size (ESS) using Tracer 1.7 software (Rambaut et al., 2014) and accepting ESS values of 200 or more.

We calculated the genetic distances among the MOTUs obtained through the three species delimitation methods using the K2P model with MEGA 7.0.26 (Kumar et al., 2016). We used the K2P, since this model allows us to compare the values found here with those previously reported in other Astyanax studies (e.g., Carvalho et al., 2011; Pereira et al., 2011; Rossini et al., 2018). Collins et al., (2012) tested whether the K2P is a well-fitted model at the species level by comparing it to the other models (JC, F81, TrN, HKY, HKY+C and GTR+C) using data sets from different animal groups, including fish. The results indicate that the differences in distance between K2P and other models were usually minimal, and the identification success rates were largely unaffected by model choice, even when interspecific threshold values were reassessed.

Multilocus analyses. To obtain a multilocus Bayesian species tree (ST) we considered the nominal species recognized by morphological studies and the results generated by the GMYC analysis. This analysis was performed in BEAST (Star-BEAST) using 500 million MCMC, sampling every 10,000, relaxed clock and Yule models, and a burn-in of 20%. Nucleotide substitution models were selected based on BIC (Bayesian Information Criterion) using jModeltest 2 (Darriba et al., 2012). The best-fitting models were HKY for COI and Cytb, and F81 + G for S7. All generations were sampled from the stationary phase. The convergence of analyses and adequate ESS (>200) were evaluated in Tracer v1.7 (Rambaut et al., 2014).

The Bayesian ST was used as guide tree to the Bayesian species delimitation approach using multilocus data (Yang, Rannala, 2010; Rannala, Yang, 2013) in the BP&P 3.3 software (Yang, 2015). This software uses a coalescent model, calculating the posterior probability of potential species considering the coalescent process of lineage sorting. The basic model used by BP&P involves two types of parameters: the population sizes on the species tree (θs), and the species divergence times (τs). To evaluate the impact of these parameters on species delimitation results and consider a range of speciation histories, we tested different gamma prior configurations and some default parameters in four distinct combinations. A first one assumed relatively large ancestral population sizes and deep divergences (θ ~ G(1,10) and τ₀ ~ G(1, 10)) among species; a second combination considered small ancestral population sizes and shallow divergences among species (θ ~ G(2, 2000) and τ₀ ~ G(2, 2000)); and the other two combinations assumed either large ancestral populations sizes (θ ~ G(1, 10)) and relatively shallow divergences among species (τ₀ ~ G(2, 2000)), or small ancestral population sizes and deep divergences (θ ~ G(2, 2000), τ₀ ~ G(1, 10)).

From the 76 individuals sampled, we obtained 75 COI sequences, comprised of 614 bp, without stop-codons, deletions, or insertions. Our primary Bayesian tree, obtained with a total of 1,867 COI sequences (75 generate herein and 1,792 downloaded from BOLD), grouped all Astyanax species belonging to the upper Paraguaçu River basin in a clade named hereafter Clade 1, with 0.95 probability posterior value (Fig. 2A), except A. hamatilis. This latter species was joined with species from other hydrographic basins (Astyanax taeniatus Jenyns, 1842 from the Ribeira da Terra Firme River; A. burgerai Zanata & Camelier, 2009 from the Almada River; Astyanax sp. from the Marcanaí River, and Astyanax sp. from the Macacuá River) in a distant clade from Clade 1.

The Clade 1 recovered 19 nominal Astyanax species from 17 hydrographic basins, which are part of the Brazilian crystalline shield and the Atlantic coast drainages (Figs. 2A, B), representing 390 sequences for A. bifasciatus Garavello & Sampaio, 2010, A. bockmanni Vari & Castro, 2007, A. aff. bockmanni, A. dissimilis Garavello & Sampaio, 2010, A. fasciatus Cuvier, 1819, A. gymnodontus Eigenmann, 1911, A. gymnogenys Eigenmann, 1911, and A. minor Garavello & Sampaio, 2010 from Paraná River basin; A. paranae Eigenmann, 1914 (Paraná and Paraguay basins); A. intermedius Eigenmann, 1908 from Paraíba do Sul basin; A. aff. intermedius from Paraná and Paraíba do Sul basins; A. cf. fasciatus and A. rivularis Lutken, 1875 from Paraná and São Francisco basins; A. scabripinnis Jennys, 1842 from Paraná, Paraíba do Sul, Paraguay, São Francisco, and Doce basins; A. bimaculatus Linnaeus, 1758 from São Francisco basin; A. laticeps Cope, 1894 and A. obscurus Hensel, 1870 from Itapocu basin; A. aff. fasciatus and A. aff. jequitinhonhae Steindachner, 1877 from Jequitinhonha basin; A. xavante Garutti & Venere, 2009 from Araguaia basin; and A. brucutu, A. epiagos, A. lorien, A. rupestris, and A. aff. rupestris from upper Paraguaçu basin (Fig. 2B).

FIGURE 2 |

Specimens of Astyanax sp. were named following the indication of the collection site reported in the BOLD database. Among the specimens identified at the genus level only (i.e., Astyanax sp.), two sequences belong to individuals from the Coité River and two belong to individuals collected in the Piabinha River, both rivers from the upper Paraguaçu River basin. Details about the samples are available in the Tab. S2.

Single locus species delimitation analyses. Our single locus delimitation analyses for the species obtained in Clade 1, with COI sequences, showed different results among the three approaches used herein (BIN, ABGD, GMYC, Fig. 2B). The BIN approach recovered five distinct MOTUs. Focusing on the species from the Paraguaçu River basin, A. brucutu, A. epiagos, and A. lorien were grouped with Astyanax sp. Coité and thirteen nominal species in a single MOTU (MOTU 1, BIN AAC5910). Astyanax rupestris (MOTU 4, BIN ADI2769) was separated from A. aff. rupestris (MOTU 5, BIN ACR6356), while Astyanax sp. Piabinha was grouped with this latter. The mean divergence within BINs ranged from 0% (MOTU 2, BIN ABZ0055) to 1.7% (MOTU 1, BIN AAC5910), and the pairwise divergence between BINs ranged from 1.8% (MOTU 4, BIN ADI2769 and MOTU 5, BIN ACR6356) to 3.6% (MOTU 4, BIN ADI2769 and MOTU 3, BIN ABZ6219).

The ABGD analysis indicated the presence of 19 distinct MOTUs into Clade 1, with four singletons, i.e., four MOTUs represented only by a single individual. The average genetic distances within and between these MOTUs were 0.18% and 2.5%, respectively. We found A. brucutu, A. lorien, and Astyanax sp. Coité grouped into the MOTU 1 with 12 nominal species. The species A. epiagos and A. rupestris were separately recovered in the MOTU 4 and 17, respectively. The genetic distance between MOTU 1 and 4 was 1.8%. Differently from the BIN analysis, A. aff. rupestris was divided in two MOTUs named A. aff. rupestris 1 (MOTU 18) and A. aff. rupestris 2 (MOTU 19), with a genetic distance between them equal to 0.7%.

The GMYC results showed a total of 50 MOTUs, of which eight were singletons. The confidence limit for the estimated number of entities ranged from 49 to 60. The null model likelihood (L₀ = 3877.946) was significantly (p < 0.01) lower than the GMYC model likelihood (L = 4039.65), indicating that there is probably more than one species in our sample. We observed A. brucutu and A. lorien grouped in the same MOTU (MOTU 26), while Astyanax sp. Coité River was clustered to the species A. bimaculatus, A. cf. fasciatus, and A. fasciatus from Miriri and São Francisco basins (MOTU 25). On the other hand, A. epiagos (MOTU 18), A. rupestris (MOTU 48), A. aff. rupestris 1 (MOTU 49), and A. aff. rupestris 2 (MOTU 50) were recovered as independent MOTUs. The average genetic distance values were 0.14% for intra- and 1.8% for inter-MOTUs.

The average genetic distance values calculated between Astyanax from the Paraguaçu River basin, defined by the BP&P analysis, were 0.0% (intra-MOTU) and 2.1% (inter-MOTU). The maximum intra-MOTU distance was 0.001% (A. rupestris), and the minimum inter-MOTU distance was 0.3% (A. brucutu and A. lorien). Astyanax aff. rupestris 1 and A. aff. rupestris 2 showed 0.7% inter-MOTU distance, while both diverged 1.8% from A. rupestris (see Tab. S3).

Multilocus species delimitation. After edition and alignment, the dataset for A. brucutu, A. epiagos, A. lorien, A. rupestris, and A. aff. rupestris consisted of 140 S7 sequences with 684 bp, and 75 COI and 73 Cytb sequences, comprising fragments with 614 bp and 1,032 bp, respectively. No stop-codons, deletions or insertions were observed.

The BEAST* analyses used to obtain a guide tree reached apparent convergence, with ESS of at least 300 for all parameters, showing convergence between runs. The BP&P results, using this prior information, separated A. brucutu, A. epiagos, A. lorien, and A. rupestris, and, similarly to the ABGD and GMYC results, suggested the existence of two genetic lineages within A. aff. rupestris (A. aff. rupestris 1 and A. aff. rupestris 2). The speciation probabilities assumed maximum values (1.0) on all nodes. Moreover, the species delimitation results were not affected by different prior settings, and we recovered maximum speciation probability values for all internal nodes in all tested combinations, indicating consistent results among runs (Fig. 3).

FIGURE 3 |

Our major phylogenetic analysis recovered all endemic species from the upper Paraguaçu River studied here in a single and large clade (Clade 1). It is noteworthy that although we have included four of the six endemic species, we did not include all Astyanax species described for the Paraguaçu River basin (Santos, Caramaschi, 2007, 2011), therefore, Clade 1 must be incomplete. Despite this, our findings suggest that fishes of this basin share an evolutionary history that can result in its high level of endemism (Buckup, 2011; Camelier, Zanata, 2014a; de Pinna et al., 2018). Interestingly, we observed a pattern of genetic proximity between species of Clade 1 and species from distinct hydrographic basins, such as the Brazilian crystalline shield and the Atlantic coastal drainages (i.e., São Francisco, Paraná, and Paraíba do Sul basins). According to Ribeiro, (2006), geological events between upland crystalline drainages and Atlantic tributaries occurred at different times, causing seemingly distant basins to share species or even species complex. The Paraguaçu River has an extensive system of branching headwaters that are adjacent to the eastern streams of the São Francisco basin (Buckup, 2011). In fact, several sister taxa or genetically related species between São Francisco River and NMAF ecoregion, presently separated by the Espinhaço Mountains, have been already reported (e.g., Camelier, Zanata, 2014a; Sarmento-Soares et al., 2016; Ramirez et al., 2017; Anjos et al., 2020).

In turn, although the species delimitation methods can be delimiting lineages, but not necessarily species (Carstens et al., 2013; Sukumaran, Knowles, 2017), our results are in accordance with the taxa considered valid to the upper Paraguaçu basin (A. brucutu, A. epiagos, A. lorien, A. rupestris) and revealed the existence of two genetic lineages within A. aff. rupestris. However, the number of identified MOTUs was method-dependent, as previously reported in similar studies (e.g., Costa-Silva et al., 2015; Rossini et al., 2016; Machado et al., 2018). These discrepancies may likely be due to analytical differences inherent to each method. The BIN and ABGD methods are based on genetic distances. The former is a result of the refined single linkage (RESL), which associates COI sequences with an identifier (BIN) based on a distance value automatically delineated (Ratnasingham, Hebert, 2013), while ABGD requires a priori specification of an intraspecific distance threshold (Puillandre et al., 2012). Contrastingly, GMYC uses coalescence approaches, and it requires an ultrametric gene tree in which branches are assigned to one lineage per species or multiple lineages per species (Pons et al., 2006). In addition, inconsistencies between the species delimitation methods may be biased by the molecular markers used (Hebert et al., 2003) and the limited sample sizes (Carstens et al., 2013) common to many investigations. It has been suggested that since distance methods rely heavily on the disparity between intra- and interspecific variation, an incomplete taxonomic sampling could influence the accuracy of the method (Frézal, Leblois, 2008).

The BIN analysis showed more conservative results, grouping the nominal species A. brucutu, A. epiagos, and A. lorien, plus Astyanax sp. from Coité River, in a single MOTU; and separating A. rupestris from the A. aff. rupestris and Astyanax sp. Piabinha. The BIN method uses 2.2% threshold, splitting species in new BINs when this value is at least twice higher (e.g., 4.4%) (Ratnasingham, Hebert, 2013). Although genetic distances equal or higher than 2% are commonly used to separate MOTUs (Hebert et al., 2004; Ward, 2009), this threshold can underestimate the number of species when applied to complex groups such as Astyanax. For Neotropical fishes, smaller genetic divergence values have been reported for congeneric species with recent divergence (e.g., Carvalho et al., 2011; Pereira et al., 2011; Ramirez, Galetti, 2015; Machado et al., 2016; Ramirez et al., 2017; Ribolli et al., 2021), and 1% has been acclaimed as the optimal threshold for those belonging to species complexes (Hubert et al., 2008; Pereira et al., 2011). Therefore, the nature of the algorithms used by the BIN method may suffer interference when employed in hyper-diverse groups. For these latter groups, GMYC analyses have been considered more efficient than other methods (Ratnasingham, Hebert, 2013; Costa-Silva et al., 2015; Ribolli et al., 2021), consisting in one of the most accepted approaches for species delimitation based on a single locus analysis (Costa-Silva et al., 2015). In Astyanax, this method was already chosen as the most appropriate for species delimitation (Rossini et al., 2016).

Following our single locus analysis and GMYC results, this study showed five MOTUs among the focused species in the Paraguaçu basin, recognizing A. epiagos (MOTU 28) and A. rupestris (MOTU 48) as distinct species, but joining the two nominal species A. lorien and A. brucutu in a single clade (MOTU 26). Of note, the multilocus species delimitation approach used here was able to separate the latter species into different MOTUs. Moreover, the description of both species was based on strong diagnostic morphological characters and distinct habitats (Zanata et al., 2017, 2018; Vita et al., 2020). According to these authors, Astyanax brucutu presents a unique mandibular morphology similarly found only in specimens of Creagrutus Günther, 1864 and Piabina Reinhardt, 1867. Furthermore, A. brucutu inhabits a geographical region characterized by a distinctive combination of environmental attributes, such as high transparent water, elevated levels of dissolved oxygen, patches of gastropod shells on the bottom and coarse substrate partially covered by aquatic macrophytes, which are not observed elsewhere in the basin or adjacent drainages (Zanata et al., 2017). Additionally, these nominal species could be easily identified by the monophyly criterion and their allopatric distribution.

Although it is parsimonious to consider A. lorien and A. brucutu as valid species, an alternative hypothesis is to assume that the morphological differences among them are possibly related to local adaptations, since the presence of barriers to gene flow can promote such phenotypic differences in distinct populations or lineages (Zamudio et al., 2016). Rossini et al., (2016) argued that Astyanax local populations described as new species due only to their restricted geographical distribution or local adaptations could be synonymized in the future. Therefore, further phylogeographic studies incorporating a larger sampling, genomic data, and considering population size and divergence time as relevant parameters should be performed to reassess the genetic relationships between A. lorien and A. brucutu.

Meantime, low genetic distances among species have been often associated with recent divergences, in which the time to accumulate genetic differences is quite short (Ornelas-García et al., 2008). Previous studies using the COI gene have already detected low genetic distances among species of Astyanax, reporting 0.93% between A. cf. fasciatus and A. rivularis (e.g., Carvalho et al., 2011). Low interspecific genetic distance values related to recent divergence have also been described within other fish genera, such as Parodon Valenciennes, 1849 (0.4%; Bellafronte et al., 2013), Zungaro Bleeker, 1858 (0.4%; Pires et al., 2017), Megaleporinus Ramirez, Birindelli & Galetti Jr., 2017 (0.67%; Ramirez et al., 2017), Leporinus Agassiz, 1829 (0.7%; Silva-Santos et al., 2018) Rineloricaria Bleeker, 1862 (0.8%; Costa-Silva et al., 2015), Apareiodon (0.9%; Bellafronte et al., 2013), and Laemolyta Cope, 1872 (0.9%; Ramirez, Galetti Jr., 2015). The small body size of specimens of Astyanax and the fact of some species are geographically isolated in headwaters may enable the occurrence of vicariance events and speciation by geographic isolation (Castro, 1999). Low genetic divergence between isolated species in distinct tributaries in the same basin may indicate that they had been through a recent vicariance followed by a fast morphological differentiation, without reaching a reciprocal monophyly (Costa-Silva et al., 2015). That might explain the low genetic divergence (0.3%) between A. lorien and A. brucutu herein observed.

The multilocus analysis, separating A. lorien from the remaining Astyanax species studied, including A. brucutu (Fig. 3), reinforces the taxonomic validity of these species. On the other hand, A. rupestris and A. aff. rupestris showed higher values of genetic distances (1.8%; see Tab. S3), suggesting that A. aff. rupestris may be indeed considered distinct from A. rupestris. This result is in accordance with the difficulties pointed by Zanata et al., (2018) in the description of A. rupestris, in which the authors decided not to include the population of the Piabinha River within A. rupestris. According to the authors, specimens of A. aff. rupestris are very similar morphologically to A. rupestris but possess variations in some meristic characters that were not observed in the former. The Piabinha population also presents high frequency of specimens with variable lateral-line perforation, four premaxillary teeth in the inner row, and reduction in the number of branched dorsal- and pelvic-fin rays (Zanata et al., 2018). Astyanax aff. rupestris is apparently restrict to the Piabinha River, a Cumbuca’s tributary, while A. rupestris is known to occurs in a somewhat broader distribution throughout both Cumbuca and Piaba River sub-basins (Zanata et al., 2018).

Furthermore, the GMYC approach also pointed A. aff. rupestris divided in two MOTUs (A. aff. rupestris 1 and A. aff. rupestris 2), evidencing hidden genetic diversity and showing MOTUs in sympatry and reciprocal monophyly (see Fig. 3), though with less than 1% of genetic distance between them (0.7%). It appears that some Astyanax lineages from the upper Paraguaçu are in a gray zone (sensu de Queiroz, 2007), in which speciation is in process, and the boundaries among species are hardly identified (Costa-Silva et al., 2015; Anjos et al., 2020). In this sense, we agree that further population studies of A. rupestris, A. aff. rupestris 1, and A. aff. rupestris 2, using methods of integrative taxonomy, including molecular and morphological data, are necessary to clarify the taxonomic status of the A. rupestris putative species complex.

Our study was useful in confirming A. rupestris from the Piaba and Cumbuca sub-basins as a single molecular unit distinct from A. aff. rupestris from the Piabinha River. In addition, the data supported the existence of two genetic lineages within the A. aff. rupestris morphotype. The multilocus analysis was more efficient in identifying species with recent divergence when compared to the single locus analysis using COI sequence only. Altogether, we characterized six distinct MOTUs: Astyanax epiagos, A. brucutu, A. lorien, A. rupestris, A. aff. rupestris 1, and A. aff. rupestris 2. Regarding the two Astyanax sp. previously reported for the Paraguaçu River basin by Rossini et al., (2016), the results indicated that Astyanax sp. from the Piabinha River and A. aff. rupestris 2 share the same COI haplotype, and, consequently, belonging to the same taxon (MOTU). On the other hand, Astyanax sp. from the Coité tributary needs to be taxonomically assessed, since it clustered to nominal species from São Francisco and Miriri basins, showing no genetic similarity to the endemic Astyanax species from the Paraguaçu River studied here. Overall, these findings contribute to a better understanding of the diversity of this fish group in the upper Paraguaçu River basin, pointing out hidden diversity and reinforcing the relevance of this hydrographic system for the biodiversity ichthyofauna.