Delphinidae

The grouping of Sotalia and Steno was strongly supported, as well as the positioning of Sousa within the Delphininae. These results indicate that Steno is the sister group of Sotalia, thus supporting the revised subfamily Stenoninae as proposed by LeDuc et al. [10]. Although those authors could not reach a definitive conclusion regarding the sister group relationship between Steno and Sotalia based on their cytochrome b data, they decided to maintain both genera in Stenoninae as indicated by earlier morphological studies. However, they suggested that the revised Stenoninae does not include Sousa, a genus also traditionally assigned to that subfamily based on morphology. Instead, they placed Sousa in Delphininae. Our phylogeny reinforces the view that Sousa belongs in the Delphininae, as formerly indicated by other studies [5], [19], [20], [21], [25], [28].

The close affinity between Steno and Sotalia and the placement of Sousa in the Delphininae were previously suggested using cytochrome b data [5], [31] and a supermatrix of nine nuclear and six mitochondrial genes [20]. Three studies disputed this conclusion. Caballero et al. [19], and McGowen et al. [21], [22] proposed placing both Sotalia and Sousa in the Delphininae, and grouping Steno with Globicephalinae+Grampus+Orcaella. In those studies, conclusions were based on a combined nuclear and mitochondrial dataset. In the case of Caballero et al. [19] and McGowen et al. [21], [22], the combined phylogeny seemed to be driven mainly by their nuclear data, since their mitochondrial phylogeny did not support such groupings. Interestingly, however, Steeman et al. [20] also combined nuclear and mitochondrial data, and recovered the same topology observed here and in the studies mentioned above. In a recent mitogenomic phylogeny of the Delphinidae Steno was also placed outside the Globicephalinae [28].

The new conformation of Stenoninae implies that the phylogenetic placement of the fossil genus Astadelphis requires revision. Astadelphis has been assigned to this subfamily, and is only recorded from Pliocene deposits (3.1–3.8 Ma) of Italy. It has been considered by different authors as phylogenetically close to Steno and Sotalia [32] or to Sousa [33]. If that latter view is correct, Astadelphis does not belong in Stenoninae.

Our phylogeny confirms the phylogenetic position of Grampus within the Globicephalinae, as first proposed by LeDuc et al. [10] and contrary to the traditional view, based on morphology, which included Grampus in the subfamily Delphininae [34]. Previous studies based on the cytochrome b [5], mitochondrial genomes [28], [35] and combined mitochondrial and nuclear genes [19], [20], [21], [22], [36] also recovered the placement of Grampus in the Globicephalinae.

On the other hand, Orcinus, traditionally placed in the Globicephalinae [34], usually figures in molecular phylogenies either as “incertae sedis” [10], [20], [21], [36], or pooled with Orcaella [5]. Recently, however, a mitogenomic study proposed the inclusion of both Orcinus and Orcaella in the Globicephalinae [28], in agreement with early views based on morphology. Our phylogeny strongly supports the exclusion of Orcinus from Globicephalinae. We also refute the existence of the subfamily Orcininae (Orcinus+Orcaella), as proposed by LeDuc et al. [10] and Agnarsson and May-Collado [5] based on cytochrome b data. Instead, Orcaella may be part of the Globicephalinae, or warrant a separate subfamily, as proposed by Perrin [34]. The relationship of Orcinus to the other delphinid subfamilies remains unresolved.

Lagenorhynchus albirostris occupies the most basal position within the delphinids. Unfortunately, since our phylogeny lacks mitogenomes from the genera Lissodelphis, Cephalorhynchus and the other Lagenorhynchus species, it is not yet possible to ascertain the evolutionary relationships of all Delphinidae subfamilies, nor the monophyly of Lissodelphininae, which has been questioned [10], [37].

Unsurprisingly, our analyses were unable to shed light on the Stenella-Delphinus-Tursiops complex, due to the current lack of mitochondrial genomes from many species. Our phylogeny only reinforced previous findings concerning the para- or polyphyly of Stenella and Tursiops [5], [10], [20], [21]. The recent speciation (<4 Ma) of these lineages poses many difficulties in phylogenetic inference (such as low number of informative characters, incomplete lineage sorting and the possible existence of fertile hybrids), and understanding their evolution may require the aid of phylogeographical approaches.

Timing of Sotalia speciation

Dating the divergence between riverine and marine Sotalia is crucial to understand the phylogeography of S. fluviatilis, as it indicates when this species became genetically isolated after colonising the Amazon basin. To date, all divergence estimates have been based either on the mitochondrial control region [23], cytochrome b [20], [21] or on both markers [18], [24]. Thus, our mitogenomic phylogeny provides the best opportunity so far to date more precisely the divergence between S. guianensis and S. fluviatilis.

Four previous studies attempted to estimate the timing of Sotalia speciation. The genetic divergence (p distance) between the Sotalia species observed by Cunha et al. [18] for both the control region and the cytochrome b was 2.5%. Taking into consideration the evolutionary rates of these markers in cetaceans – control region: 0.5% to 1% per million years [38]; cytochrome b: 1%/Ma, [39] – the speciation event that separated both lineages would have happened between 2.5 and 1.25 Ma, during the early Pleistocene [24]. A somewhat similar estimate was obtained using a relaxed molecular clock and cytochrome b data, 1.99 Ma (0.63–3.67) [21]. The dates proposed by Cunha et al. [24] and McGowen et al. [21] overlap with our estimate – 2.3 Ma (1.3–3.4). A different timing of the speciation was proposed by Caballero et al. [23], who calibrated a molecular clock for the control region using the estimated divergence between Sotalia and Phocoena phocoena based on the fossil record (10 to 11 Ma). Therefore, they arrived at a faster substitution rate, and dated the divergence between S. fluviatilis and S. guianensis much later, at 1.0 to 1.2 Ma. Finally, the oldest time estimate for the separation of Sotalia species (3.5 Ma) was obtained by Steeman et al. [20]. It is noteworthy that, in spite of being a supermatrix analysis (of 15 mitochondrial and nuclear markers) using seven fossil calibration points and relaxed clock models, the divergence between the Sotalia species in that study was based exclusively on the cytochrome b gene (the only marker analysed from S. guianensis by the authors).

Besides being the most precise estimate available, our dating coincides remarkably well with the establishment of the modern Amazon River basin. Until recently, authors accepted that the modern Amazon basin was already established in the Miocene, but that view has changed based on new geological data. Sediment analyses showed that the Amazon River only attained its present conformation by the beginning of the Pleistocene, approximately 2.5 Ma [40], [41]. At the same time, there was a major lowering of sea level from 3 to 2 Ma [42], which could have been partly responsible for changing the river's course eastwards, coupled with Andean tectonics [40], [41]. Irrespective of the environmental conditions prevailing at the time, Sotalia dolphins that colonised the Amazon basin certainly had an Atlantic origin, because the connection with the Caribbean via the Paleo-Orinoco river and the Paleo-Maracaibo had been closed since the rising of the northern Andes cordillera, 8 Ma [43], [44].

DNA extraction, amplification and sequencing

Total DNA was isolated from skin samples of Steno bredanensis, Sotalia guianensis and S. fluviatilis using the standard phenol-chloroform procedure [45]. Two long fragments (about 9 Mb and 7 Mb), comprising the entire mitochondrial genome, were PCR-amplified using the primers described by Sasaki et al. [46]. Amplifications were carried out in 50 µL reactions using the Qiagen LongRange PCR Kit. Reagent concentrations and cycling profile followed the manufacturer's instructions.

Long-PCR products were then used as templates for amplification of smaller fragments, using the primers described in Xiong et al. [25] and others developed in this study (Table S1). PCR reactions (30 µL) contained 1.5 U Taq, 200 µM dNTP, 2.5 mM MgCl₂, 15 µg BSA and 0.5 µM of each primer. Amplification thermal conditions were as follows: 3 min at 93°C, 30 cycles of 1 min at 92°C, 1 min at 50°C or 55°C, 1 min at 72°C, and 5 min of final extension at 72°C. PCR products were purified using ExoSap (GE) and both strands were sequenced in an ABI3130 using BigDye chemistry.

Sequences were edited in SeqMan 7 (DNAStar), using the complete mitochondrial genome of Sousa chinensis [GenBank EU557091] as template to build the contigs. The complete mitogenomes were deposited in GenBank (JF681038, JF681039 and JF681040).

Alignment and evolutionary analyses

All cetacean mitochondrial genomes available in GenBank as of March 2011 were included in this study (Table S2). Alignments were conducted for each gene individually in ClustalW [47] and manually checked. A supermatrix of 15,873 bp was used in phylogeny estimation and divergence time inference. It consisted of the concatenation of all 13 protein coding genes, tRNA and rRNA genes and D-loop. Protein coding genes were further separated into three partitions containing only first, second and third codon positions. This was done to maximise rate variation among partitions [48]. Phylogeny estimation was performed in MrBayes 3 [49], PhyML 3 [50] and RAxML 7.0.4 [51]. In MrBayes and RAxML, each partition was allowed to evolve independently under the GTR+G+I model.

For the Bayesian inference, the Markov chain Monte Carlo (MCMC) settings used were as follows. Two independent runs, with three Markov chains each, were sampled every 100^th generation during 10⁷ generations, resulting in 100,000 trees in each run, of which 25% were discarded as burn-in. In RAxML, maximum likelihood (ML) topology estimation was conducted independently from 200 different starting trees. Then, 1,000 bootstrap pseudoreplicates were run to obtain the statistical support for the nodes of the tree with the highest log-likelihood. ML tree search in PhyML was performed by the SPR algorithm and the aLRT statistic [52] was used to evaluate node confidence.

Divergence time inference was conducted in BEAST 1.6.1 [53] using the same data partitioning, substitution model and MCMC settings described above. Substitution rate evolution was modelled by the uncorrelated lognormal distribution. Tree topology prior followed the Yule process.

Calibration information used as priors

The ages of five nodes were constrained by calibration information based on the fossil record of cetaceans (Figure 3):

1. The time since the most recent common ancestor (TMRCA) of Cetacea was calibrated by a gamma prior with shape = 1.0, scale = 4.8 and offset = 33.5 Ma. This is based on the early Mysticeti fossils at the Eocene/Oligocene boundary [54], [55]. The gamma prior was adjusted so that the tail of the distribution would include the Archaeoceti fossils, which are supposedly the stem cetacean lineage.
2. The earliest members of the Odontoceti [56] were found in the late Oligocene [57]. By around 23.7 Ma, odontocetes had already diversified, because this is the age of Ferecetotherium, which presents autapomorphies of the Physeteridae [16]. We have used a gamma prior with shape = 1, scale = 4.5 and offset = 23 Ma. The tail of the gamma distribution was extended to the Eocene, to safely incorporate stem lineages of the mysticetes and odontocetes.
3. The age of the Monodon/Phocoena split was constrained by a gamma prior with shape = 1, scale = 2 and offset = 10.5 Ma. Prior information was based on the oldest fossil Phocoenidae, Salumiphocaena stocktoni, from the late Miocene of North America [58]. The shape of the distribution was set in order to accommodate the Miocene epoch.
4. The diversification of Delphinidae was constrained by a normal prior with mean = 10.5 and standard deviation = 1.0 Ma, which resulted in a 95% interval from 8.5 to 12.5. This prior was based on the record of early delphinids from the late Miocene [59].
5. The divergence between Iniidae and Pontoporiidae had already taken place in the early Pliocene. We used a gamma prior with shape = 1.7, scale = 2.5 and offset = 5 Ma to estimate the TMRCA of Inia/Pontoporia. Xiong et al. [25] used the fossil Brachydelphis to constrain the Inia/Pontoporia divergence. However, the extensive morphological revision by Geisler and Sanders [16] places Brachydelphis as stem lineage of “Platanistoidea”, which includes extant Lipotes and Platanista as well as Inia and Pontoporia. Thus, we chose not to consider this fossil as a stem Pontoporiidae and established an offset for the gamma prior at 5 Ma, which safely includes late Miocene South American fossils of iniids and pontoporiids [60].

Download:

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

Figure 3. Calibration information used as priors in Bayesian dating analyses.

(a) TMRCA of modern Cetacea; (b) TMRCA of Odontoceti; (c) Age of the Monodon/Phocoena split; (d) Age of the Delphinidae diversification; (e) Age of the Iniidae/Pontoporidae divergence.

https://doi.org/10.1371/journal.pone.0028297.g003