Field Collection

Blood samples of about 50 μl were withdrawn from each individual via femoral vein puncture with syringes containing an NE buffer (NaCl 100 mM and EDTA 10 mM, pH 8) as an anticoagulant and then preserved in tubes with absolute ethyl alcohol, labeled with a field number, and stored in a refrigerator (4°C) until processing. Individuals were released in the same areas where they had been captured. The blood samples were deposited in the Animal Genetic Sample Collection of the Laboratório de Evolução e Genética Animal of the Universidade Federal do Amazonas (UFAM), Brazil. The collecting sites (Fig. 1) were 10 localities from throughout the known distribution of P. erythrocephala in Brazil: 1) São Gabriel da Cachoeira (Negro River; n = 18), 2) Santa Isabel do Rio Negro (Ayuanã River; n = 34), 3) Cumicuri River (n = 29), 4) Barcelos (Itu River; n = 29), 5) Viruá National Park (Iruá Creek; n = 10), 6) Jaú National Park (Jaú River; n = 34), 7) Nhamundá (Paracutu River; n = 23), 8) Barreirinha (Andirá River; n = 33), 9) Terra Santa (Jamary Creek, community Alema; n = 17), and 10) Oriximiná (Sapucuá Lake; n = 19).

Laboratory Methods

Genomic DNA was isolated from blood by dissolving and digesting samples with a Proteinase K/SDS solution, followed by successive washes in phenol, phenol/chloroform/isoamyl alcohol, and hydrated chloroform; a subsequent addition of 5 M NaCl; and a final addition of 70% ethyl alcohol for DNA precipitation (Sambrook et al. 1989). The mitochondrial control region was amplified via polymerase chain reaction (PCR) using primers PRO (5′-CCCATCACCCACTCCCAAAGC-3′; Sites et al. 1999) and the conserved primer 12SR5 (5′-GGCGGATACTTGCATGT-3′; Hrbek and Farias 2008). For the amplification of each sample, we used 0.8 μl of DNA (approximately 30 ng), 2 μl of each 2-μM primer (PRO and 12SR5), 2.5 μl of buffer (200 mM Tris-KCl, pH 8.5), 1.5 μl of 25-mM MgCl₂, 2.5 μl of 10-mM dNTP, 0.2 μl of Taq polymerase enzyme (5 U/μl), and 13.5 μl of ddH₂O in a final reaction volume of 25 μl. The PCR reactions were performed under the following cycling conditions: 35 cycles consisting of 1 min at 92°C (denaturation), 35 sec at 55°C (annealing of primers), 1 min at 72°C (primer extension), and final extension at 72°C for 5 min. The PCR products were subjected to purification by precipitation with sodium acetate (3 M, pH 4.8) and 95% ethanol.

The sequence reactions were performed in a final reaction volume of 10 μl using the DYEnamic ET dye terminator kit (GE-Healthcare, São Paulo, Brazil). Each reaction had the following composition: 4 μl of amplified DNA, 2 μl of 2-μM primer, 2 μl of ET-terminator, and 2 μl of ddH₂O. The samples were then subjected to a 35-cycle thermocycling protocol with the following temperature profile: 20 sec at 95°C (denaturation), 15 sec at 55°C (annealing of primers), and 1 min at 60°C (primer extension). At the end of the PCR reaction, samples were purified according to manufacturer's instructions (GE-Healthcare). Subsequently, the samples were subjected to electro-injection and the sequences were resolved on an automatic sequencer MegaBACE 1000 DNA Analysis System (GE-Healthcare, São Paulo, Brazil) following the manufacturer's instructions.

Descriptive Statistics

The sequences resolved on the automated sequencer were imported and manually verified in the program BioEdit (Hall 1999). Alignment was carried out in BioEdit using the program ClustalW (Thompson et al. 1996) under default conditions. Genetic diversity was estimated from gene diversity (Ĥ; Nei 1987), nucleotide diversity (Π; Nei and Li 1979), the number of haplotypes of the mtDNA, and the number of segregating sites (S). These measurements of genetic variability were estimated with the programs Arlequin v3.5 (Excoffier and Lischer 2010) and DNAsp v5 (Librado and Rozas 2009). Relationships among haplotypes were estimated within a network of haplotypes based on a maximum likelihood phylogenetic tree topology using the program HAPLOVIEWER (Salzburger et al. 2011).

Population Structure and Gene Flow Estimates

Population structure was assessed using the analysis of molecular variance (AMOVA; Excoffier et al. 1992) to test the hypotheses that rapids are barriers to gene flow and that the sediment-laden Amazon River is a barrier to gene flow for the clearwater- and blackwater-inhabiting irapuca. The AMOVA; the Φ_ST, which is analogous to F_ST (Weir and Cockerham 1984); and the Mantel test (Mantel 1967) were implemented in the software Arlequin v3.5 (Excoffier and Lischer 2010). Population structure was assessed with the Bayesian analysis of population structure in BAPS v5.2 (Corander et al. 2003), which infers the most likely number of genetic clusters within an analyzed sample. The program BAPS treats the nucleotide frequencies and the predefined number of clusters as random variables and jointly calculates the posterior distributions of the population structure and allele frequencies. Uncertainty of the unknown population structure (i.e., the probability that allele frequencies are statistically not different from each other) is estimated for all pairwise cluster comparisons, and final posterior likelihood(s) of population structure is estimated.

An approximate estimate of gene flow (Nm) between sampled localities was calculated from Φ_ST under the island-migration model, in which Φ_ST = 1/(Nm + 1). However, considering that real populations are likely to violate many of the assumptions of the above relationship (Whitlock and McCauley 1999), we also estimated female-mediated gene flow using the maximum likelihood framework in the program MIGRATE v2.4.4 (Beerli and Felsenstein 2001). We ran 10 short chains, sampling each chain 10,000 times, and then 6 long chains, sampling each chain 1,100,000 times and discarding the first 100,000 samples as burn-in.

The hypothesis of isolation-by-distance was tested by the Mantel test (Mantel 1967). Values of Φ_ST/(1 − Φ_ST) were correlated with river distances in kilometers between each pair of sampling localities (Fetzner and Crandall 2003). The significance of the Spearman correlation (Spearman 1904) was tested with 10,000 permutations in the software Arlequin v3.5 (Excoffier and Lischer 2010).

Demographic History

Population demography was estimated by mismatch distribution in 2 different ways: in the program DnaSP v5 (Librado and Rozas 2009), where the distributions of pairwise differences are compared with an expected distribution under a null model of a stationary population, and in the program Arlequin v3.5 (Excoffier and Lischer 2010), which estimates the sums-of-squared deviation and Harpending's raggedness index (Harpending 1994), with significance tested through 10,000 permutations under the null model of population expansion. Tajima's D (Tajima 1989) and Fu's Fs (Fu 1997) tests were used to verify that the samples were at mutation–migration–drift equilibrium. Significant values for these tests could indicate that the mitochondrial sequences are not evolving according to the hypothesis of selective neutrality (i.e., they are not in a mutation–drift equilibrium) or that the populations were previously subdivided and/or experienced fluctuations in the past (i.e., they are not in migration–drift equilibrium; Hartl and Clark 2006). The significance of both statistics was tested by comparisons of the statistics against a distribution generated from 10,000 random samples on the assumptions of selective neutrality and population equilibrium. The tests were implemented with the program Arlequin v3.5 (Excoffier and Lischer 2010).

We estimated the female effective population sizes (Nef since we are dealing with an mtDNA marker) based on coalescence theory using Watterson's estimate of the population genetic parameter θ from s (Watterson 1975) and assuming a mutation rate of 2.48 × 10⁻⁷ mutations per site per year for the control region; this estimate was derived from Chelonia mydas (Chassin-Noria et al. 2004). The genetic parameter θ was derived from the analysis implemented in the program MIGRATE v2.4.4 (Beerli and Felsenstein 2001).

Genetic Diversity and Demography

The aim of this study was to assess genetic diversity and its distribution among populations of P. erythrocephala and to test whether 1) rapids and waterfalls are effective geographic barriers to gene flow and 2) the main channel of the Amazon River is as an effective barrier to gene flow. Average gene diversity (Ĥ = 0.608 ± 0.037) and nucleotide diversity (Π = 0.00217) observed in P. erythrocephala were similar to values observed by Pearse et al. (2006) in P. expansa (Ĥ = 0.65 ± 0.305; Π = 0.00256). However, both gene and nucleotide diversity were highly heterogeneous among the sampled localities (Table 1). The haplotype network showed a common haplotype found in most of the localities; however, groups of haplotypes separated by more than 1 mutational step from the common haplotype were observed for São Gabriel da Cachoeira (locality 1), Barcelos (locality 4), Jaú National Park (locality 6), and Barreirinha (locality 8), each of which showed a certain level of geographic structuring in relation to the other localities (Fig. 2).

The genetic diversity indices and the significantly negative values in the Tajima's D and Fu's Fs statistical tests (p < 0.05) observed for individuals from the Cumicuri River (locality 3) indicate that this population may be undergoing or has already undergone recent population expansion. The Barcelos (locality 4), Nhamundá (locality 7), Barreirinha (locality 8), and Oriximiná (locality 10) populations did not present significant values for the Tajima's D test, but Fu's Fs test was significant. Both tests are negative when there is an excess of recent mutations, and negative values are interpreted as evidence of population growth and/or diversifying selection (Tajima 1989; Fu 1997). According to (Ramos-Onsins and Rozas 2002), the Fs statistic is more sensitive to detection of population expansion and takes into account the influence of sample size in the analysis. Interestingly, the majority of samples that showed population expansion are located in the Negro River basin. Despite the fact that the neutrality tests were significant when all localities were analyzed together, the results of demographic tests (mismatch distribution) of individual localities and all localities pooled together suggest that P. erythrocephala is in mutation–drift equilibrium. The lack of equilibrium observed in the neutrality tests when all localities are analyzed together is an artifact of the analysis and indicates population structure (Hartl and Clark 2006), as already observed in the haplotype network (Fig. 2) and BAPS analyses (Fig. 3). In summary, it is unlikely that the species as a whole or several individual populations are experiencing demographic growth.

Population Structure and Connectivity

Based on our analyses, P. erythrocephala from most of the sampled localities belong to a large contiguous population occurring in the central Brazilian Amazon. No isolation by distance was observed within this main population. However, the differences showed by Φ_ST values for some pairwise comparisons (Table 2) and the rejection of panmixia by AMOVA suggest that some localities are genetically differentiated (Φ_ST = 0.27931). These genetically differentiated populations are present in the upper Negro River upstream of São Gabriel da Cachoeira (locality 1) and in the Jaú River in Jaú National Park (locality 6). Both localities are separated by rapids or small waterfalls. The Andirá River locality near the Barreirinha municipality (locality 8) is separated from the others by the main channel Amazon River and thus isolated by the physicochemical properties of this river. The genetic differentiation of these 3 localities lends considerable support to the hypothesis of geomorphological and physicochemical barriers to gene flow and the need to identify appropriate management units (MUs). Once those localities that are geographically separated by geological barriers are removed from analyses of population structure, the absence of population structuring of the remaining localities becomes apparent.

There is no correlation between pairwise Φ_ST and geographic distances between sampling localities; thus, the hypothesis of isolation by distance is not supported. Lack of isolation by distance has also been reported in P. expansa and P. unifilis (Bock et al. 2001) and P. lewyana (Vargas-Ramírez et al. 2012), although the lack of isolation by distance may be an artifact of low sampling density in both of these studies. However, 2 other studies using denser sampling and microsatellite markers suggest a possible influence of geographic distance on the pattern of genetic variability in P. expansa (Pearse et al. 2006) and P. unifilis (Escalona et al. 2009) populations.

Waterfalls and Rapids as Barriers to Gene Flow

São Gabriel da Cachoeira (Negro River, locality 1) and Jaú National Park (Jaú River, locality 6) are separated by rapids and a small waterfall, respectively, from all other populations of P. erythrocephala. Both localities show differentiation considering Φ_ST values and are the primary contributors to the rejection of panmixia in AMOVA. This genetic differentiation, detected by 3 different methods of analysis—AMOVA, Φ_ST, and BAPS—reinforces the importance of the waterfalls and rapids in the Negro River basin in genetic structuring of P. erythrocephala.

In the upper and middle course of the Negro River, there are numerous geomorphological structures, such as waterfalls and rapids. Some waterfalls are permanent, with steeper gradients, while others present gentler slopes and become submerged during the high-water period. The river's water level, which ranges from 6 to 8 m, depends on the season. During the low-water season, the waterfalls and rapids emerge, hindering or preventing the movement of some aquatic species, while in the high-water season, currents are much stronger and may impede movement of some aquatic animals. An area of rapids separates the locality upstream of São Gabriel da Cachoeira (locality 1) and the Jaú River locality (locality 6).

The pattern of geographic structuring reported in P. erythrocephala was similar to that observed in P. expansa in the Madeira River basin (Pearse et al. 2006). In their study, Pearse et al. (2006), using control region sequences, observed that 2 populations sampled from the Madeira-Guaporé River clustered closely together and shared a private haplotype that was not found outside the Madeira-Guaporé River basin. These 2 populations are located in the upper Madeira River above the 290-km stretch of rapids (Cella-Ribeiro et al. 2013), which was then hypothesized to act as a substantial barrier to the gene flow in P. expansa. Nevertheless, these localities also share 2 more frequent haplotypes with other populations sampled in the Amazon region, suggesting a relatively recent structuring of these populations. In the same study, Pearse et al. (2006), using microsatellite markers, found population differentiation of P. expansa individuals above the falls and rapids present in the upper Madeira River and middle Tocantins-Araguaia River.

In addition to structuring populations of P. erythrocephala and P. expansa, rapids and waterfalls in the Amazon basin also structure fish species (Farias et al. 2010; Ochoa et al. 2015), crocodilians (Hrbek et al. 2008), and dolphins (Gravena et al. 2014, 2015; Hrbek et al. 2014). Rapids also structure entire fish communities (Torrente-Vilara et al. 2011). All of these examples are from the Madeira and in the case of Hrbek et al. (2014) also from the Tocantins-Araguaia River, where the effect of rapids is best studied. Studies of the effects of other rapids on aquatic fauna are effectively nonexistent, making the P. erythrocephala study one of the first studies focusing on the Negro River basin.

The Amazon River as a Geographic Barrier to Gene Flow

The riverine hypothesis states that the large rivers in the Amazon basin isolate populations on opposite banks, thus limiting gene flow and allowing populations to diverge via selection or genetic drift (Haffer 2008). Genetic evidence fully or partially supporting this hypothesis is limited to the studies of terrestrial vertebrates, such as nonvolant birds (Ribas et al. 2012), frogs (Gascon et al. 1998; Fouquet et al. 2015), and lizards (Oliveira et al. 2016). There is also evidence that rivers delimit small birds (Fernandes et al. 2012, 2014; Fernandes 2013; Weir et al. 2015). It might seem paradoxical that large rivers would delimit the distribution of aquatic organisms. However, P. erythrocephala is restricted to clearwater and blackwater environments (Mittermeier et al. 2015) and prefers small water bodies over main river channels (Rueda-Almonacid et al. 2007). Therefore, our results offer support for the hypothesis that major rivers can act as barriers to movement and gene flow for P. erythrocephala. The Andirá River locality near the Barreirinha municipality (locality 8) is the only sample from a southern tributary of the Amazon River and thus is isolated from all other localities by the main channel and the physicochemical properties of the Amazon River. Pairwise Φ_ST data also showed that individuals from Andirá River locality were genetically differentiated in relation to other localities, except when the comparisons involved the populations of Jamary Creek (locality 9) in the Terra Santa municipality and Iruá Creek (locality 5) in Viruá National Park. While reasons for the lack of significant differentiation between the Andirá River and Jamary Creek localities seem clear, it is unclear why no significant geographic differentiation should be observed between the Andirá River and Viruá National Park.

Despite the results presented in here, the genetic differentiation observed between P. erythrocephala from the Andirá River locality and most of the other localities should be viewed cautiously since it may also be the result of an ecological process, local adaptation, or some historical event that has resulted in the differentiation of the population of the Andirá River locality.

Implications for Conservation and Management

Recent studies demonstrate that molecular data can also be used as an important component in the conservation of threatened species, as they allow the identification of population structure and subsequently of evolutionary significant units and MUs (Moritz 1994; Crandall et al. 2000; Fraser and Bernatchez 2001; Palsbøll et al. 2007). The identification of biological populations and knowing the geographic scale at which populations are demographically connected or independent are important for species monitoring and management (Moritz 1994; Palsbøll et al. 2007). From a conservation perspective, São Gabriel da Cachoeira (locality 1), Jaú National Park (locality 6), Barreirinha (locality 8), and all remaining sampling localities represent 4 distinct MUs based on the criteria of Moritz (1994), which must be administrated as separate entities, given that they comprise genetically distinct stocks.

As the São Gabriel da Cachoeira (locality 1), Jaú National Park (locality 6), and Barreirinha (locality 8) localities demonstrate high levels of genetic differentiation from the remaining centrally located population and are geographically restricted, we suggest that for any future conservation measures, the protection of only a small number of nesting beaches and the transfer of individuals between different areas would not be an appropriate management strategy for this species. Multiple protected nesting beaches and boiadouros (the places where females concentrate before and during the spawning period) are a priority and need to be located in the vicinity of each MU. Furthermore, any future transfer or translocation of individuals (hatchlings) needs to be done within each respective MU. With respect to the fourth MU representing the main Amazonian group, any future translocation of individuals can be done over a larger geographic area, albeit still within the geographic distribution of the MU. Future translocation plans could be valuable conservation and management actions in cases of future severe local population decline, such as in areas of the Negro River basin where indiscriminate hunting of adult females and collection of eggs is common.

While our study clearly identifies 4 MUs based on the criteria of Moritz (1994), conservation and management decisions are complex, and many other sources of information must enter into the decision-making process. Other biological sources of information, including from the nuclear genome, which may show different patterns of differentiation than the mitochondrial genome (Palsbøll et al. 2007), and the ecological role of the populations (Hunter and Hutchinson 1994) should be considered. Socioeconomic factors are also important in conservation. There are a sufficient number of highly polymorphic microsatellite loci for species of the genus Podocnemis (Pearse et al. 2006; Fantin et al. 2007) to follow up this study to refine our hypotheses about gene flow and population structuring, which, together with demographic and ecological data (Batistella and Vogt 2008), may be used to better delimit the number and distribution of distinct MUs of P. erythrocephala.

Podocnemis erythrocephala is currently considered vulnerable; however, there are no data on the effect of exploitation on the species, there are no historical data on the rate of its consumption, and there is only anecdotal evidence that it was much more abundant around 20 yrs ago (Mittermeier et al. 2015). In conclusion, there is an urgent need for more studies of all aspects of the biology of this species. We can properly conserve and manage only what we know well.