A study of cranial shape in dolphins of genus Sotalia was done using 104 specimens (92 from localities along the Brazilian coast and 12 from the Amazon River basin). Twenty-two cranial landmarks, assumed to be homologous, were selected for analysis. The first 2 principal components of aligned coordinates explained 40.6% of the total variation in cranial shape. Although no sexual dimorphism was detected (P = 0.811), shape differences among populations of Sotalia were highly significant (P < 0.000001). The 1st and 2nd principal components of shape showed that the Sotalia population from the Amazon basin differed in cranial shape from marine populations. Based on differences in geometric shape, a discriminant analysis of 3 linear measurements between landmarks provided an equation that classified skulls as belonging to Amazonian or marine populations. Based on these results and evidence from several other divergent character systems and life history attributes, we suggest the use of Sotalia guianensis for marine dolphins and S. fluviatilis for Amazonian dolphins.
The genus Sotalia (Cetacea: Delphinidae) was described in 1866 by Gray (see Hershkovitz 1966), based on a skull from French Guyana, and the specimen was named Sotalia guianensis (see van Bénéden 1864 in Hershkovitz 1966). About the same time, 2 new species were described, with S. fluviatilis (see Gervais and Deville 1853 in Hershkovitz 1966; =S. pallida =S. tucuxi) applied to dolphins from the Amazon basin (northern Brazil), and S. brasiliensis van Bénéden 1875 to dolphins restricted to Guanabara bay (Baía de Guanabara), Rio de Janeiro State, southeastern Brazil (Hershkovitz 1966).
These species names and areas of occurrence were accepted until reports confirmed the presence of Sotalia along the coast of several Brazilian states and in other Central and South American countries (e.g., Bossenecker 1978; Carvalho 1963; da Silva and Best 1996; Husson 1978; Simões-Lopes 1988). Sotalia is now known to have an extensive distribution along the Atlantic coast of Central and South America and in the Amazon (da Silva and Best 1994, 1996). These dolphins are commonly observed inshore, in rivers, bays, and estuaries, where they follow the original distribution of mangroves, from Honduras (da Silva and Best 1996) to the state of Santa Catarina in southern Brazil (Simões-Lopes 1988).
The external morphology of marine and freshwater Sotalia is similar, although there are differences in their habitat usage. Based on these observations, Rice (1977) suggested the existence of only 1 species, S. fluviatilis, with 2 subspecies: S. f. fluviatilis in the Amazon and S. f. guianensis in marine waters. Based on a study of 59 skulls from marine and freshwater dolphins, Borobia (1989) subsequently strengthened the argument for only 1 species. In this case, conventional multivariate morphometrics confirmed the size variation noticed before, but the scores for specimens of the 2 populations overlapped considerably in the second principal component (shape component) and could explain only 6% of the total variation in the distance measurements of skulls.
With the acceptance of Sotalia as a monotypic genus, some authors (Best and da Silva 1984; da Silva and Best 1994, 1996) have used information obtained from dolphins in the Amazon to make generalizations about various aspects of the biology of Sotalia, mainly because the tucuxi (S. f. fluviatilis) is the most studied of the Sotalia dolphins (Alcuri and Busnell 1989; Best and da Silva 1984; Caldwell and Caldwell 1970; da Silva 1990; da Silva and Best 1996; Harrison and Brownell 1971; Magnusson et al. 1980; Zam et al. 1970). Such generalizations suggest that morphological evidence for the assignment of taxa in Sotalia requires reexamination using more powerful, recently developed methods.
Considering that morphological structures in an organism have 2 components (size and shape), and that shape is the most informative for defining biological entities in nature (Atchley et al. 1992; Patton and Brylski 1987), we evaluate in this study whether the component of shape differences between Amazonian and marine Sotalia is really negligible, as has been suggested by multivariate analysis of distance measurements and, therefore, insufficient to justify the recognition of more than a single biological entity. To address this question, we applied the recently developed method of geometric morphometrics (Bookstein 1991; Kendall 1984; Rohlf 1996, 1998), which allows the analysis of geometrical shape in biological structures.
Materials and Methods
The 104 specimens (Appendix I) used were from collections held by the Laboratório de Mamíferos Aquáticos, Universidade Federal de Santa Catarina (LAMAQ/UFSC), Universidade Federal do Paraná (UFPR), Instituto de Pesquisas Cananéia (IPeC), Museu de Zoologia, Universidade de São Paulo (MZUSP), Museu de História Natural, Universidade Estadual de Campinas (ZUEC), Museu Nacional (MN/UFRJ), and Projeto Mamirauá. The specimens were grouped based on geographic origin as marine south (n = 32), marine southeast (n = 47), marine northeast (n = 13), and Amazonian (n = 12).
Twenty-two landmarks (Fig. 1), assumed to be homologous in all specimens, were defined for the dolphin skulls as follows: 1 = rostral tip, 2 and 12 = anteriormost point of the notch in the maxilla, 3 and 11 = intersection between the frontal bone and zygomatic process, 4 and 10 = intersection between the parietal bone and frontal–interparietal suture, 5 and 9 = posteriormost point on the curve of the parietal, 6 and 8 = posteriormost point on the curve of the occipital condyle, 7 = posteriormost point on the edge of the supraoccipital, 13 = midpoint of the nasal bone suture, 14 = anteriormost point of the suture between the frontal and interparietal bones, 15 and 22 = dorsalmost point on the pterygoidal notch, 16 and 21 = point in the suture between the frontal and alisphenoid bones, 17 and 20 = ventralmost point of the basioccipital crest, and 18 and 19 = ventralmost point of the paraoccipital process.
The skulls were photographed in 3 views (dorsal, lateral, and ventral) at an angle of 90° from each other (dorsal and ventral views were 180° apart). The pictures were digitized with a flatbed scanner and superimposed with the constraint that landmarks 1 and 7 were at the same coordinate position in the 3 pictures of each specimen (these 2 landmarks could be identified in all views). The 2-dimensional coordinates were combined in such a way that the y-dimension in the lateral view (the dorsoventral axis) corresponded to the z-dimension in the dorsal and ventral data sets. This combination produced a 3-dimensional data set of skull landmarks. Skull size was calculated as the centroid size of landmark configurations (Bookstein 1991) which corresponded to the square root of the summed squared distances between each landmark and the configuration centroid (average point in each landmark constellation).
The landmark configurations were superimposed by an orthogonal least squares technique known as Procrustes superimposition (Rohlf and Slice 1990; Slice 1996). This technique translates, scales, and rotates the landmark configurations so as to minimize the summed squared distances between corresponding landmarks. In the generalized least squares superimposition (GLS), a mean shape is iteratively estimated and all specimens in the sample are superimposed on the mean.
In 3-dimensional objects, 7 nuisance parameters are removed from the coordinates after superimposition: 3 axes of translation, 1 scale (centroid size), and 3 angles of rotation. Although this information is removed, the set of coordinates retains the same number of dimensions (3p, where p = number of landmarks), whereas the real dimensionality of the data set should be 3p − 7. One way to solve this problem is to reduce the dimensionality of the set of superimposed landmark coordinates using a multivariate technique (Dryden and Mardia 1998). We used principal component analysis to remove the extra dimensions from the aligned coordinates (the last 7 principal components have eigenvalues equal to 0) and to search for biologically interesting phenomena in the directions of major variation within shape space. The eigenvectors can be depicted in figure space as estimated shapes (icons) with high scores in both positive and negative directions (Dryden and Mardia 1998). The principal components of superimposed coordinates are equivalent to other variance-maximizing rotations of shape variables commonly used in morphometric studies, such as relative warp analysis (Rohlf 1999).
We used the first 2 principal components of the Sotalia sample, with sexes and geographic origin combined as shape variables in multivariate analyses of variance (MANOVA). As there were too many skulls with no information on sex, the interaction between sex and geographic origin could not be evaluated, and the analyses had to be performed separately for each factor, instead of using a 2-way model. To test for sexual dimorphism and geographic effects on size, we performed analyses of variance (ANOVA) on centroid size, classifying the samples by sex and locality.
To assess the relative importance of localized versus global shape changes, we performed a generalized affine least squares superimposition (GALS). This technique superimposes the landmark configurations after removing the nuisance parameters from the coordinates and stretches and compresses the configurations in orthogonal directions (affine transformations). The number of dimensions after the affine superimposition is smaller than those of after the orthogonal superimposition. For 3-dimensional objects, there are 3p − 12 dimensions (the 7 nuisance parameters plus 5 uniform dimensions). The coordinates resulting from this superimposition were used in a 2nd principal components analysis. The difference between the ordinations obtained using the results of orthogonal superimposition and the affine superimposition reveals the effect of discarding uniform components and of performing the analysis in a localized shape subspace (Bookstein 1996).
The first 2 principal components of the superimposed coordinates (orthogonal least squares) explained 40.6% of the total shape variation in the sample. The MANOVA performed on the 2 principal components classified on the basis of sex was not significant (Wilks' lambda = 0.98458, d.f. = 2, 27, P = 0.810748), whereas that performed on the same components classified on the basis of geographic origin (marine south, marine southeast, marine northeast, and Amazonian) was highly significant (Wilks' lambda = 0.093936, d.f. = 6, 198, P < 0.0000001). The ANOVA for sexual dimorphism on centroid size was not significant (F = 0.253083, d.f. = 1, 128, P = 0.618847), whereas the ANOVA for geographic effects on size was highly significant (F = 38.56772, d.f. = 3, 100, P < 0.0000001). As no sexual dimorphism in skull shape or size was observed (either because there was none or because the information available was incomplete), we analyzed only geographic variation.
The scatterplot for the first 2 principal components (Fig. 2) from GLS showed 2 groups of samples, marine and Amazonian. The 1st principal component depicted a major shape difference between the Amazonian and marine specimens. Morphologically, there were differences in shape between the 2 populations (Fig. 3). In Amazonian specimens, the rostrum and the occipital condyle pointed downwards relative to the anteroposterior axis of the skull, whereas in marine specimens the rostrum and the condyle were aligned along this axis. Consequently, the skulls of Amazonian specimens look arched with the concavity on the ventral side; the region of the zygomatic process was relatively larger in marine specimens than in Amazonian ones. The braincase was relatively larger in marine than in Amazonian specimens, and the distance between the anterior notch of the maxilla (landmarks 2 and 12) and the nasal bone (landmark 13) was relatively greater in Amazonian specimens.
The ordination of specimens on the first 2 principal components from the GALS (results not shown) was similar to that obtained with orthogonally superimposed coordinates. The first 2 principal components from GALS accounted for about the same percentage of sample variation as the GLS principal components (41%), but the latter concentrated more variation in the 1st axis.
The scatterplot between the 1st principal component from GLS and centroid size (Fig. 4) showed an allometric pattern between Amazonian and marine populations. However, this allometric pattern between groups was not an extrapolation of the within-group pattern. An analysis of covariance of the scores of the 1st principal component from GLS, classified by locale and using centroid size as a covariate, was still significant (F = 103.047, d.f. = 3, 99, P < 0.0000001, test of parallelism: P = 0.995).
A discriminant analysis (classifying by 2 groups: marine and Amazonian) was done on linear measurements between landmarks (obtained from the 3-dimensional coordinates) in the skulls. We used 3 distances that, judging from the results of the geometric analysis, were expected to reflect the shape differences between the 2 populations: distance between landmarks 1 and 7 (D1,7; basicranial axis length), distance between landmarks 1 and 13 (D1,13; tip of the snout to nasal bone), and distance between landmarks 16 and 21 (D1,16; skull width).
To avoid confusion in the analysis by the large size differences between groups, we related each distance to basicranial axis length using the ratios (D1,13)/(D1,7) and (D16,21)/(D1,7) as variables. As expected, the discriminant analysis using the two ratios was highly significant (Wilks' lambda = 0.407675, F = 73.37318, d.f. = 2, 101, P < 0.000001), and the marine specimens had higher scores than the Amazonian ones on the discriminant axis. The percentage of correct classifications was very high. For the marine specimens, only 2 out of 92 were incorrectly classified. The entire sample of 12 Amazonian specimens was correctly classified. The structural coefficients for the ratios on the discriminant axis were: (D1,13)/(D1,7) = −0.80625 and (D16,21)/(D1,7) = 0.59578. Thus, the allometric pattern for discriminant structural coefficients indicated that, although the marine specimens were larger than Amazonian specimens, the former had a relatively larger skull width at the base and a relatively smaller length from the tip of the snout to the nasal bone (all relative to skull length). The classification functions obtained were 1002
The 2 scores were calculated for each specimen, and specimens were then grouped according to their highest classification scores. Using these functions, it was possible to identify whether a Sotalia skull belonged to a marine or Amazonian specimen based only on 3 skull measurements. These classification functions may be useful for researchers who use museum specimens of unknown geographic origin.
Our results provide new insights into the pattern of skull shape and size variation in the genus Sotalia. The only factor that influences shape divergence among specimens was geographic origin. The lack of sexual dimorphism in the skull dimensions, although not confirmed here because of lack of information on the sex of some specimens, has been observed earlier in this group (Borobia 1989).
Like other small cetaceans, such as Stenella (Perrin et al. 1981, 1987, 1991) and Delphinus (Evans 1994; Perrin et al. 1985), the genus Sotalia is a subject of taxonomic controversy at species level. Initially, 8 species of Sotalia were described (which included Sotalia and Sousa). A century and a half later only 1 species is accepted, although doubts remain, and subspecific names have been attributed to Amazonian (S. fluviatilis fluviatilis) and marine (S. f. guianensis) dolphins (cf. Rice 1977). Borobia (1989), who recognized only 1 species based on a conventional multivariate morphometric analysis of linear distances, suggested that Amazonian and marine specimens differed only in size.
Geographic differences in size have been reported earlier in Sotalia (Mitchell 1975) and do not support the creation of taxonomic boundaries between closely related organisms (Atchley et al. 1992; Patton and Brylski 1987). In contrast, shape variation is a more reliable and richer source of information about biological processes. When defined as the geometric properties of an object that are invariant to the effects of rotation, translation, and scaling (Dryden and Mardia 1998), the shape of biological structures is best studied using geometric methods that are unaffected by these confounding effects. Traditional morphometric methods, such as principal components of distance measurements, are confounded by size differences among specimens (Rohlf and Bookstein 1987) and cannot be reliably used to determine shape differences between biological entities.
The 1st principal component from the generalized least squares superimposition showed a large amount of shape variation between marine and Amazonian specimens. There was also considerable overlap among marine specimens of different geographic origins (south, southeast, and northeast), indicating that these specimens are morphologically uniform. The same result was obtained for the first 2 principal components of the GALS, indicating that differences in global shape were not important for discerning between the two groups of specimens. On the other hand, the description of shape differences uncovered large- and small-scale localized shape differences between the two groups.
The differences in shape between the skulls of marine and Amazonian dolphins were large. The alignment of the rostrum and occipital condyle differed markedly between the 2 groups of dolphins. In marine dolphins, the opening of the foramen magnum is located further posterior, whereas in Amazonian dolphins this opening is directed downwards. This variation in the position of the foramen magnum suggests that in marine dolphins the cranium would be in line with the vertebral column, whereas in Amazonian dolphins the cranium would point downwards.
This difference in skull shape may reflect a functional distinction between marine and Amazonian dolphins. The downward inflection of skulls in Amazonian dolphins may be associated more with the need to scan river beds which are frequently littered with tree trunks and branches that might cause dolphins to strand, rather than with the type of food taken, as only 11% of the fish consumed are bottom-dwellers (da Silva 1983). On the other hand, for marine dolphins, about 60% of the fish and crustaceans (e.g., shrimps) consumed are bottom-dwellers (M. R. Oliveira, in litt.), and diving toward the bottom during fishing activities is common (Monteiro-Filho 1992, 1995), with no risk of stranding.
When 2 groups of closely related organisms differ in size and shape, it is important to establish whether the differences in shape reflect heterochronic phenomena alone, such as hypermorphosis (Alberch et al. 1979), in which the 2 groups of organisms share a common linear growth trajectory but with 1 group attaining a larger size and, consequently, a different shape. One could argue that the relatively larger braincase and smaller skull size seen in Amazonian specimens are evidence that these dolphins have retained early ontogenetic features of shape and size (neotenic differentiation), or that marine dolphins attain a larger size and different shape (hypermorphosis).
Examination of the size and shape vectors for the skulls of marine and Amazonian Sotalia indicates that, although the 2 groups differ in size and shape, the within-group major axes of variation are not linearly related (i.e., they do not correspond to simple linear extensions of each other), as would be expected if skull shape differentiation in these groups could be attributed only to hypermorphosis. On the other hand, the pattern of size and shape differentiation between the two groups suggests a static intraspecific allometric pattern (Cheverud 1982) that has no simple heterochronic basis. A thorough study on skull shape differentiation between Amazonian and marine Sotalia and its ontogenetic basis would require the inclusion of young individuals in the sample, but young specimens are uncommon in museum samples.
The geometric descriptors and multivariate statistical analyses used here revealed a substantial morphological discontinuity in cranial shape between Amazonian and marine dolphin populations. There are, in addition, several other character systems in which the Amazonian and marine groups differ markedly: size of adults, with smaller dolphins living in river systems and larger ones living in the sea (Borobia 1989; Mitchell 1975); social organization, with a polyandrous system for Amazonian dolphins (Best and da Silva 1984) and a family organization for marine dolphins (Monteiro-Filho 2000); presence of only 1 (left) functional ovary and a gestation time of 10–10.3 months in Amazonian dolphins (Best and da Silva 1984; Harrison and Brownell 1971) compared with both ovaries being functional and a gestation time of 11.3–12 months in marine dolphins (Rosas and Barreto, in press); seasonal testicular activity in Amazonian dolphins and continuous testicular activity in marine dolphins (F. Rosas, in litt.); significant differences between the mitochondrial DNA sequences of marine and Amazonian populations (M. Furtado-Neto, in litt.); and dissimilar echolocation clicks between the 2 populations (Kamminga et al. 1993).
This evidence strongly suggests that morphogenetic mechanisms and evolutionary processes have acted independently on marine and Amazonian dolphins, leading to the currently observed distinct patterns of morphological, ecological, and behavioral traits. We consider this evidence to indicate that Sotalia is not monotypic, and suggest the use of separate names to designate the 2 groups currently recognized as Sotalia fluviatilis. Based on priority, the names should be Sotalia guianensis for marine dolphins and S. fluviatilis for Amazonian dolphins.
Um estudo da forma craniana em golfinhos do gênero Sotalia foi realizado utilizando 104 espécimens (92 oriundos de localidades costeiras e 12 da Bacia do Rio Amazonas), sendo utilizados vinte e dois marcos anatômicos considerados homólogos. Os 2 primeiros componentes principais explicaram 40.6% da variação de forma craniana. Não foi detectado dimorfismo sexual (P = 0.811), entretanto a diferença de forma entre as populações foi altamente significativa (P < 0.000001). Os dois primeiros componentes de forma do crânio mostraram que os Sotalia da Bacia Amazônica diferem da população marinha. Baseado nas diferenças de forma geométrica, uma análise discriminante de 3 medidas lineares gerou uma equação que classifica os crânios como sendo da Bacia Amazônica e da população marinha. Com base nesses resultados e em outras características da história natural destes animais, nós sugerimos o uso de Sotalia guianensis para os golinhos marinhos e S. fluviatilis para os golfinhos da Bacia Amazônica.
We thank M. Marmontel (Sociedade Civil Mamirauá), J. A. de Oliveira, L. B. F. Oliveira, and L. O. Salles (Museu Nacional, Rio de Janeiro), M. de Vivo (Museu de Zoologia, Universidade de São Paulo, São Paulo), and P. C. Simões-Lopes (Laboratório de Mamíferos Aquáticos, Universidade Federal de Santa Catarina) for allowing access to specimens, and M. Furtado-Neto, F. C. W. Rosas and M. R. Oliveira for their observations. We thank F. C. W. Rosas and K. D. K. A. Monteiro who provided critical and insightful comments, J. R. Somera who helped with the line drawings, E. P. Lessa who provided thoughtful comments that greatly improved the clarity of the manuscript, and S. Hyslop for reviewing the English of the manuscript. This work was funded by the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq), Fundação de Amparo à Pesquisa do Estado de São Paulo (grant number 99/06845-3), and Instituto de Pesquisas Cananéia. ELAMF and SFR are partially supported by research fellowships from CNPq. Work by LRM is funded by the Fundação Estadual do Norte Fluminense.
The 104 specimens used were obtained from the following collections: Laboratório de Mamíferos Aquáticos, Universidade Federal de Santa Catarina (LAMAQ/ UFSC: 1073; 1079; 1082; 1083; 1087; 1104; 1108; 1114; 1117; 1130; 1175; 1180; 1203; 1208; 1218; 1219; 1222; 1226), Universidade Federal do Paraná (UFPR: 011; 014; 020; 027; 029; 031; 033; 052; 113; 114; 134),Instituto de Pesquisas Cananéia (IPeC:008; 010; 012; 013), Museu de Zoologia, Universidade de São Paulo (MZUSP: 9417; 9611; 9821; 10230; 10232; 10403; 18874; 18923; 18943; 18944; 18948; 18949; 19541; 19913; 23801; 23802; 23809; 23810; 23811; 23812; 23813; 24811; 24812; 26852; 26853; 26855; 26856; 26857; 26858; 26859; 26860; 26863; 26866; 26867; 26868; 26870; 26871; 27521; 27522; 27523; 27560; 27561; 27591; 27592; 27653; 27654; 27830; 27831; 27997; 27998; 27999; 28000; 28181; 28182; 28183; 28184;), Museu Nacional (MN/UFRJ: 001; 004; 009; 010; 014; 124), and Projeto Mamirauá (EEMSF 9502; 9507; 9508; 9509; 9510; 9514; RDSMSF 97; 97/2; and 1 specimen not numbered). The specimens were grouped by geographic origin as marine south (n = 32), marine southeast (n = 47), marine northeast (n = 13), and Amazonian (n = 12).