Geographic Variation in Skull Shape of the Water Rat Scapteromys Tumidus (cricetidae, Sigmodontinae): Isolation-by-distance plus Environmental and Geographic Barrier Effects?

The geographic variation in skull size and shape of the swamp rat Scapteromys tumidus was examined in samples from eight geographic clusters in almost of its distribution in southern Brazil and Uruguay. For analysis we used two-dimensional geometric morphometric methods for dorsal, ventral and lateral views of the skull. The geometric descriptors showed no signifi cant differences in skull size between geographic clusters, while differences in shape were highly signifi cant. We found a signifi cant and moderate correlation between geographic and morphological distances, corroborating the isolation-by-distance model. Samples from the Rio Grande do Sul central coastal plain were the most differentiated, segregating completely from all other samples in canonical variate analysis for the dorsal view. The most visible variable regions in skull were the zygomatic arch (mainly the squamosal root of zygomatic) and the lateral braincase borders. Once correlation between geographic and morphological distances were not strong, it is possible that other factors (environmental heterogeneity and/or geographic barriers) may are acting in S. tumidus skull differentiation.


INTRODUCTION
Microevolution is a process of small evolutionary changes within populations of a determined species (Avise 2000).Such changes in genetic and morphological characters within and among populations characterize different evolutionary units over geographic space (dos Reis et al. 2002a).
In relation to morphological traits, intraspecific variations occur as a result of genetic diversity, selection, reproductive isolation and other factors.Microevolutionary events, however, are the driving force of macroevolution (evolution beyond species level) processes, resulting in speciation (Avise 2000).
Intraspecific morphological variation can be accessed by using methods of traditional and geometric morphometric analysis (Rohlf and FERNANDO M. QUINTELA, RODRIGO FORNEL and THALES R.O.FREITAS Marcus 1993, Fruciano et al. 2011).The traditional or multivariate methods are based on analysis of distances (i.e.lengths and widths), angles and ratios, while the geometric morphometrics techniques capture the shape of structures in two-dimensional (2-D) or three-dimensional (3-D) coordinates of landmark points (Rohlf andMarcus 1993, Adams et al. 2004).Therefore, geometric morphometrics represents an advance over traditional methods, since visualization of shape variation is possible through graphic representations (Rohlf andMarcus 1993, Monteiro anddos Reis 1999).In addition, morphometric geometrics also deals with the size factor by the properties of the centroid size (Bookstein 1991), a vector independent from shape (Monteiro and dos Reis 1999).
The detection and description of geographic variation on skull size and shape are important for investigations on mammalian systematics and taxonomy (Barčiová 2009, Yazdi 2011).In relation to rodents, geographically based studies have addressed closely related species (i.e.Fadda and Corti 1998, 2000, 2001, Beolchini and Corti 2004, Mullin et al. 2004a, b, Barčiová and Macholán 2006) or variations within single species (dos Reis et al. 2002a, Fornel et al. 2010, Yazdi et al. 2011).In the latter cases, distinct morphological geographic units were revealed, concording with the isolation-by-distance model, in which an increase on geographic distance results in a decrease of gene fl ow, increase of genetic divergence and consequently higher morphological differentiation (Wright 1943).
The morphological traits of Scapteromys have been researched, which resulted in taxonomic and systematic reevaluations.Gyldenstolpe (1932) redescribed the genus, providing new morphological observations and listing five species (S. aquaticus, S. fronto, S. gnambiquarae, S. tomentosus, and S. tumidus).Ellerman (1941) provided generic external and craniodental anatomical defi nitions and listed the fi ve named forms recognized by Gyldenstolpe (1932).Massoia and Fornes (1964) compared qualitative (coat color, fronto-parietal suture and mesopteygoid fossa shape) and quantitative (external, cranial, penile and bacular) characters between Argentinean and Uruguayan samples, assuming a subespecific classifi cation for (S.t. tumidus for Uruguay and S. t. aquaticus for Argentina).Hershkovitz (1966) presented external, craniodental, penile and bacular anatomical descriptions and some external, cranial and bacular measures of S. tumidus, and considered the Argentinean and Uruguayan forms as distinct species.A signifi cant contribution to the systematics of Scapteromys was made by D' Elía and Pardiñas (2004), who analyzed 564 skulls from 68 populations in Paraguay, Argentina and Uruguay.Working with qualitative traits and craniodental morphometrics, in addition to mitochondrial DNA, these authors reinforced the existence of two clades at a specifi c level, fi nding S. aquaticus in Paraguay and Argentina plus one locality in western Uruguay, and S. tumidus in Uruguay.Despite the efforts already undertaken, Scapteromys lacks complete descriptive and integrative information, seeing that none of these studies properly analyzed the Brazilian populations.
The aim of this study was to investigate variation in skull shape and size of Scapteromys tumidus among populations sampled over most of the species distribution.Therefore, we performed statistical analysis of data generated by 2-D geometric morphometrics.We tested the correlation between geographic distance and morphological differentiation.This information was used to test the isolation-by-distance hypothesis, which might affect skull morphological traits in Scapetomys.

GEOMETRIC MORPHOMETRICS
Undamaged skulls were photographed in the dorsal, ventral and left lateral views.Damaged skulls were photographed only in the view(s) with the landmark regions intact.The photographs were taken with a digital camera with 10 megapixels (4608 × 3456) of resolution, zoom of 2.1 × and external accessory illumination.Camera lens was positioned parallel to the photographic background, by means of an adjustable support platform.Millimeter graph paper was used as a photographic background, aiding in skull alignment and subsequent scale referencing.For photographs of the dorsal and ventral views, the skulls were positioned on the background with modeling clay and the molars surface were aligned with the background with the aid of a set-square.For the photographs of the lateral view, skulls were fi xed on the background by the zygomatic arch and the molar surfaces were aligned with the horizontal lines of the millimeter paper.Twentytwo two-dimensional landmarks were collected in the dorsal view, 23 in the ventral view and 22 in the lateral view of the skull (Fig. 2, Appendix II).Coordinates of landmarks were obtained using the software TPSDig2, version 2.16 (Rohlf 2010).Coordinates were superimposed using a generalized Procrustes analysis (GPA) algorithm (Dryden and Mardia 1998), which removed the effects of scale, orientation and position differences (Rohlf and Marcus 1993).The size of each skull in each view was estimated from its centroid size, which is defi ned as the square root of the sum of the squares of the distances of each landmark from the centroid (Bookstein 1991).Because we used different distances in skull photographs, a size correction was made by IMP CoordGen6f (Sheets 2001), using a scale factor.

SIZE
The sexual dimorphism was evaluated using two groups, formed by the set of male or female specimens of all populations.Size differences between sexes for each view were tested with a Student's t-test on log-transformed centroid size.The ANOVAs were used to compare the log centroid size for geographic clusters, and for multiple comparisons we used Tukey's test.

SHAPE
The difference in shape between sexes for each view was tested with a multivariate analysis of variance (MANOVA) on shape variables.In order to improve the multivariate analyses and graphic visualization, neighboring localities without a potential barrier to gene fl ow (large rivers, lagoons and estuary mouths) were grouped in 8 geographic clusters, named: Campanha Gaúcha (CG), West margin of Mirim lagoon (WM), São José do Norte restinga (SJN), Rio Grande restinga (RG), Soriano (SO), Maldonado (MA), South of Mirim lagoon (SM), Canelones/Montevideo (CM) (Fig. 1).Because the MANOVA did not detect signifi cant differences in shape, and only detected small signifi cant differences in size for the dorsal view between sexes (see results), males and females were pooled for analysis of the geographic variation.A consensus confi guration of landmark confi gurations of all clusters was computed for the dorsal, ventral and lateral views, based on the orthogonal generalized least-squares Procrustes method (Rohlf and Slice 1990).The hypothesis of the existence of significant differences in skull shape among geographic clusters was tested with the MANOVA method for each view.Next, we performed a pairwise MANOVA, aiming to identify the pairs of clusters that showed signifi cant differences in shape.We performed a linear discriminant analysis (LDA), calculated on a subset of PC's to compute the leave-one-out cross-validation to calculate the percentages of the correct classifi cation for each clusters defi ned a priori (Baylac and Friess 2005).We performed a Canonical Variate Analysis (CVA) as a discriminant analysis, with all clusters pooled and the distribution of clusters visualized in the multivariate space of the fi rst two canonical axes.The adequacy of the isolation-by-distance model for our data was examined by two tests.The fi rst uses morphometric data to calculate the Mahalanobis distance matrix (morphological distances) between pairwise clusters.The morphological similarities and dissimilarities between clusters were visuali zed in a phenogram resulting from the Neighbor-joining method, using Mahalanobis distances for each view.The second is a Mantel test, which analyzed the correlation between the two matrices, geographic distances and Mahalanobis morphological distances, using 10 000 random permutations.The geographic distance matrix was based on linear distances of each locality constructed by the software Geographic Distance Matrix Generator, version 1.2.3 (Ersts 2009).For the statistical analyses and to generate graphs, we used the "R" language for statistical computing, version 2.14.1 (R Development Core Team; http:// www.r-project.org) and the libraries "ape" (Paradis et al. 2004), "ade4" (Dray and Dufour 2007), "MASS" (Venables and Ripley 2002), and "stats" (R Development Core Team).For geometric morphometrics, the procedures were carried out using the "Rmorph" library for R (Baylac 2008).
The MANOVA tests detected significant differences in skull shape among geographic clusters for all views (dorsal: λ Wilks = 0.0044; F (7, 172) = 3.482; P = 2.2 × 10 -16 ; ventral: λ Wilks = 0.0063; F (7, 164) = 2.793; P = 2.2 × 10 -16 ; lateral: λ Wilks = 0.0053; F (7, 159) = 2.988; P = 2.2 × 10 -16 ).The pairwise MANOVA showed signifi cant differences in 22 of 28 comparisons between geographic clusters.The F values and signifi cance levels are shown in Table I.FERNANDO M. QUINTELA, RODRIGO FORNEL and THALES R.O.FREITAS For the dorsal view, the first two canonical variate axes (CV1 and CV2) explain 45.9% (26.6% and 19.3%, respectively) of the variation in shape among geographic clusters.The projection of individual scores shows an overlap of the clusters, except for cluster SJN, which was completely separated along negative CV1. Cluster MA (topotypes) is discriminated in positive CV1 with a small overlap with the closest cluster CM.The major shape differences associated to the two fi rst axes in this view are represented in Figure 3a.Specimens from cluster SJN showed a slender rostrum, a broader zygomatic (with anteriorly positioned squamosal root), fronto-parietal suture posteriorly positioned and a less infl ated braincase.Specimens in cluster MA showed a shorter rostrum, a narrower zygomatic (posterior positioned squamosal root), fronto-parietal suture anteriorly positioned and a more infl ated braincase.For the ventral view, the first canonical variate axes explained 48.7% of the shape variation (CV1: 26.4%; CV2: 22.3%).We observed a marked overlap of the geographic clusters, with a clear discrimination of cluster SJN along the positive CV2 (Fig. 3b).Differences in shape for the two main axes for the lateral view are represented in Figure 3c.Cluster SJN specimens  were associated with a posteriorly positioned squamosal zygomatic root and slightly anterior tympanic bulla.In lateral view, the fi rst two CV axes explain 42.8% of the shape variation (CV1: 25.4%; CV2: 17.4%).The cluster SJN sample was discriminated on the negative CV1, while the Uruguayan clusters SO, MA and SM grouped on the positive CV1.Cluster SJN specimens tended to an anteriorly extended premaxillary, posterior squamosal zygomatic root, and broader braincase and tympanic bulla.The Uruguayan clusters on the positive CV1 tended to a shorter anterior premaxillary region, anterior squamosal zygomatic root, and less robust braincase and tympanic bulla.The percentages of correct classification for each view of the skull are shown in Table II.Considering the mean for the three views, cluster SJN showed the highest percentage of correct classifi cation (91.1%), while cluster WM showed the lowest percentage value (60.7%).The Mantel's test showed signifi cant correlation between the morphological and geographic matrices for the dorsal (R = 0.55, P = 0.011), ventral (R = 0.51, P = 0.015), and lateral (R = 0.47, P = 0.021) views of the skull.These results suggest that there are regular and signifi cant associations between skull shape and linear geographical distances among the groups.For the dorsal view, the greatest Mahalanobis distance was found between clusters SJN (middle Rio Grande do Sul coastal plain) and MA (Maldonado; type locality) (Fig. 4a).The phenogram topology showed a clear division of two groups, one formed by the Uruguayan cluster branches and the other formed by the Rio Grande do Sul clusters branches.For the ventral view, the greatest Mahalanobis distance was found between clusters SJN and SO (Fig. 4b).The phenogram topology did not show the geographic structure recovered from the dorsal view, with some Uruguayan and Rio Grande do Sul cluster branches in intercalated positions.For the lateral view, the greatest Mahalanobis distance was also found between clusters SJN and MA (Fig. 4c).In addition, the phenogram showed the topology of a division in two groups, one formed by the Uruguayan (South) and another formed by the Rio Grande do Sul (North) cluster branches, but with differences in the relative positions of some clusters in relation to dorsal phenogram.

DISCUSSION
We found marked geographic variation in shape of the water rat Scapteromys tumidus skull.The Mantel's test showed a significant and regular correlation between the geographic and morphological distances, which corroborates with the isolation-by-distance model.This model for population genetics predicts that the gene flow is inversely related to the increase in geographic distances, enhancing the genetic divergence between "subgroups" (Wright 1943).Interpopulational variation in skull morphology correlated with geographic distance has been found previously in echimyid (dos Reis et al. 2002a, b) and murid (Fadda and Corti 1998, 2000, Yazdi et al. 2011) rodents.
The phenogram generated from the dorsal view data (the most informative on geogra phic variation) showed spatially structured morphological relationships, where the greatest distance was between the central Rio Grande do Sul Coastal Plain samples (cluster SJN) and the topotypes from Maldonado, Uruguay (cluster MA), in the far northeast and extreme southern parts of the species' range, respectively.This geographical structure was also recovered from the lateral view data.Since geographic distances limit dispersal, the rate of migration is higher between nearby populations than between distant populations (Wright 1943, Bradburd et al. 2013).This pattern was corroborated by our data from the dorsal and lateral views, where the small morphological distances were found between nearby localities.
The segregation of cluster SJN from all other clusters in the canonical analysis for the dorsal view data is noticeable, indicating a higher differentiation of these populations.Mitochondrial DNA data revealed that S. aquaticus and S. tumidus are reciprocally monophyletic sister species (D'Elía and Pardiñas 2004, D'Elía 2003, Quintela et al. 2014).Such information, associated with an analysis of the present-day distribution of both species, leads to the hypothesis that the divergence between S. aquaticus and S. tumidus took place along the Lower Uruguay River basin and, according to Freitas et al. (1984), S. tumidus is in the process of colonization eastward of the River Plate.In this context, S. tumidus populations from Rio Grande do Sul central coastal plain, far from the species center of origin, resulted from long range dispersal events, and may have experienced a morphological differentiation due to reduced gene fl ow and genetic drift.According to Wright (1943), in "long range populations" or "subdivisions", there is a tendency to the fi xation of determined alleles, so that the differentiation within subdivisions is little, whereas the differentiation between subdivisions is marked, drawing a pattern of "local differentiations" resulted from the fixation of different alleles among populations.Additionally, in long dispersal populations, the fi xation of exclusive mutant alleles also enhances the local differentiation (Wright 1943).Therefore, it is possible that fi xed alleles unique to central Rio Grande do Sul state coastal plain populations have enhanced the morphological differentiation.Nevertheless, because the morphological and geographic distances were not strongly correlated, it is expected that other factors are also associated with the geographic variation in S. tumidus skull shape.In addition to the geographic distance, the existence of possible geographic barriers can also hamper the gene fl ow (Fernandes et al. 2009, Fornel et al. 2010).The spatial heterogeneity over the area covered by the study should also be borne in mind, seeing that the samples of S. tumidus used for the analysis came from different palustrine physiognomies (e.g.herbaceous swamps, arbustive swamps, peat and riparian restinga forests, river and stream fl oodplains).
The skull structure is presumably under high environmental pressure, because it is related to feeding and sensitive functions (Voss et al. 1990).In heterogeneous landscapes, the genetic responses of populations occur directly, via selective processes, or indirectly, via demographic (effective sizes, subdivisions) and stochastic events.These responses produce the genetic variation in species (Pamilo 1988).Subsequent phenotypic variation among populations results from selection and genetic variation, in addition to phenotypic plasticity (Adkinson 1995).Phenotypic plasticity occurs when, under distinct environmental conditions, more than one phenotype is expressed from the same genotype (Pigliucci 2001).This last factor, therefore, is also associated with spatial heterogeneity.Skull morphological differentiation in the murids Gerbillus (Dipodillus) (Abiadh et al. 2010), Mastomys (Lalis et al. 2009), and Taterillus (Dobigny et al. 2002), in the ctenomyid Ctenomys minutus (Fornel et al. 2010) and in the Japanese shrew-mole Urotrichus talpoides (Wilson 2013) were attributed, among other factors, to environmental heterogeneity.An analysis of the shape confi gurations (Fig. 3) showed that variable structures in S. tumidus skull comprise the zygomatic arch, the squamosal root of zygomatic and the braincase.The zygomatic arch is the site of attachment of some of the masticatory muscles (mandibular adductors): the masseter complex is connected along the maxillary root and median section of the arch; the anterior part of the zygomatic-mandibularis is attached to the maxillary root, while the posterior part of this complex is connected from the squamosal root to the glenoid fossa.Another muscle complex involved in mastication is the temporalis.The anterior and posterior temporalis are large muscles attached to the lateral braincase and mandible coronoid process, with the anterior temporalis passing through the posterior orbital fossa (Voss 1988).The regions visually variable in S. tumidus skull, however, are associated to attachment of the muscles involved in the process of mastication.Thus, associated to isolation-bydistance effects, it is possible that local variations on components of the habitat and natural history (e.g.type of soil, feeding habits) may also have led to skull differentiation in S. tumidus.The effect of this and other ecological variables associated to the environmental heterogeneity, however, must be properly tested for a better clarifi cation.
Geographic barriers are known to play an essential role in rodent genetic and morphological differentiation (Colombi et al. 2010, Nascimento et al. 2013).The populations analyzed in the present study are distributed in areas that are part of Precambrian to Holocene geological formations (Bossi and Navarro 1988, Philipp et al. 2000, Tomazelli and Villwock 2000).This range includes several different hydrographic elements: rivers flow into rifts of Precambrian granitic plutons; groundwater upwelling and rainfall excavated the sediments in wind-dispersed mantles of the Holocene coastal plain, which originated coastal streams of varied width; events of marine transgression-regression associated with Pleistocene glaciations and deglaciations resulted in depositional barriers during the maximum marine transgressions, which isolated the Patos-Mirim lagunar system, the largest lagunar complex in South America (Vieira 1983, 1984, Philipp et al. 2000, Tomazelli and Villwock 2000).The role of water bodies as potential geographic barriers for small mammals is much discussed (see Silva and Patton 1998, Cáceres 2007, Colombi et al. 2010, Fornel et al. 2010, Nascimento et al. 2013), especially considering species with some degree of adaptation to aquatic habits, as is the case of S. tumidus (Sierra de Soriano 1969).In view of the discrimination of the cluster SJN samples, it is conceivable that the Patos-Mirim lagunar system (specifi cally the Patos lagoon estuary, São Gonçalo channel and Mirim lagoon) may form physical barriers to gene fl ow among populations from both margins.The prevalence of comparatively high F values between pairs including cluster SJN in the pairwise MANOVA analysis is also noteworthy.Based on these dissimilarities, samples from the central Rio Grande do Sul Coastal Plain may comprise a distinct geographic morphological unit.To clarify whether this morphological discontinuity is incongruous with S. tumidus as a single taxonomic unit, com plementary data such as molecular analysis are needed.These analysis are highly recommended, in view of the marked differences between the cluster SJN samples and the topotypes (cluster MA) in dorsal and lateral views, as well as the phenogram distances discussed above.Coordenação de Aperfeiçoamento Pessoal de Nível Superior (CAPES), Fundação de Amparo a Pesquisa do Estado do Rio Grande do Sul (FAPERGS) and Conselho Nacional de Desenvolvimento Científi co e Tecnológico (CNPq).
Specimens of S. tumidus from nine localities in Rio Grande do Sul were collected from March 2003 through October 2011.Specimens from three additional localities in Rio Grande do Sul were obtained from the Museu de Ciências Naturais of Universidade Luterana do Brasil (MCNU), Canoas, Brazil.Specimens from seven localities in Uruguay and one locality in Rio Grande do Sul were photographed at American Museum of Natural History (AMNH), New York, United States.Sample sizes are given in Appendix I and sampled localities are shown in Figure 1.The material examined consisted of 180 specimens (107 males and 73 females) for the dorsal view, 172 specimens (102 males and 70 females) for the ventral view and 167 specimens (101 males and 66 females) for the lateral view.All of the specimens were adults according toBarlow (1969), with the third molar fully erupted and showing signs of wear.

Figure 1 -
Figure 1 -Map with locations of Scapteromys tumidus sampled populations and geographic clusters in southern Brazil and Uruguay.Information on voucher specimens are listed in Appendix I.

Figure 2 -
Figure 2 -Landmark locations on dorsal (a), ventral (b) and lateral (c) views of skull of Scapteromys tumidus.Detailed description of the location of landmarks is given in Appendix II.

Figure 3 -
Figure 3 -Scatter plot of two fi rst axes of Canonical Variate Analysis (CVA) for Scapteromys tumidus skull shape confi gurations of samples from eight geographic clusters in southern Brazil and Uruguay for dorsal (a), ventral (b) and lateral (c) views.CV 1 represent skull shape variation in the fi rst axis and CV 2 represent skull shape variation in the second axis.Positive scores in solid lines and negative scores in dotted lines on the shape difference representations.

Figure 4 -
Figure 4 -Neighbor-joining phenograms generated from Mahalanobis distances for dorsal (a), ventral (b) and lateral (c) views of the skull of Scapteromys tumidus.Acronyms correspond to geographic clusters.