Taxonomic identification using geometric morphometric approach and limited data : an example using the upper molars of two sympatric species of Calomys ( Cricetidae : Rodentia )

The taxonomic identification of micromammals might be complicated when the study material is fragmented, as it is the case with pellets and fossil material. On the other hand, tooth morphology generally provides accurate information for species identification. Teeth preserve notably well, retaining their original morphology, unlike skulls and mandibles, which can get crushed or have missing parts. Here, we explored a geometric morphometrics approach (GM) to identify fragmented specimens of two sympatric Calomys Waterhouse, 1837 species – Calomys tener (Winge, 1888) and Calomys expulsus (Lund, 1841) – using the morphology of intact molars as the basis for identification. Furthermore, we included some specimens of uncertain taxonomic identification to test their affinities and the utility of the shape of the molar to identify incomplete specimens. We evaluated the variations in the shape of the first upper molar (M1) among 46 owl pellets specimens of Calomys, including C. expulsus (n = 15), C. tener (n = 15), and unidentified specimens treated as Calomys sp. (n = 16) through GM analysis using 17 landmarks. The data was explored using PCA, PERMANOVA, and Discriminant analyses over the Procrustes residuals matrix were applied to evaluate interand intraspecific shape differences. Also, we evaluated whether allometric shape differences could impact the data, but found no evidence of a correlation between size and shape. Our results support that shape differences in the M1 are effective for discriminating between C. tener and C. expulsus. Moreover, the unidentified specimens do not represent a third shape but could be identified with confidence either as C. tener or C. expulsus. Our results show that even with fragmentary materials, GM is a feasible and useful tool for exploring inter-specific shape differences and assisting in taxonomic identification as a complement to traditional qualitative description of diagnostic features in poorly preserved specimens.


INTRODUCTION
The complex morphology of the molars (with the cones, flexes, and lophs) is a source of information for the study of cricetid rodent taxonomy, as these structures provide diagnostic characteristics for subfamilies (Reig 1977).Over the years, it has been demonstrated that geometric morphometrics is a useful tool for systematics, for example taxonomic identification by analysis of molar shape differences among rodents, especially of fossil material (Polly and Head 2004, Kryštufek and Janžekoviè 2005, Macholán 2006, Marcolini et al. 2009).Enamel hardness protects the molars so they often are the only intact structure of a fragmented skull and form the only known elements of several extinct taxa in the fossil record (Reig 1977).
Among the genera with complicated taxonomy and great morphological similarity between some species is Calomys Waterhouse, 1837 (Almeida et al. 2007).They are small cricetid rodents distributed mainly in areas of dry vegetation, with wide distribution in South America (Bonvicino et al. 2010).The Brazilian species can be separated into two major groups, based on skull and body size measurements: a group of larger-bodied Calomys that includes Calomys callosus (Rengger, 1830), Calomys expulsus (Lund, 1840), Calomys tocantinsi Bonvicino, Lima & Almeida, 2003, Calomys callidus (Thomas, 1916), and Calomys cerqueirai Bonvicino, Oliveira & Gentile, 2010; and a group of smaller-bodied individuals with two species, Calomys tener (Winge, 1888) and Calomys laucha (G. Fisher, 1814).Despite this morphometric clustering, the smaller body size group is not monophyletic and Calomys laucha shares a more recent common ancestor with C. expulsus than C. tener does (Almeida et al. 2007).In contrast to their general morphological and morphometric similarity, the karyotype differs greatly between species of Calomys, making cytogenetic studies jointly with molecular data useful for species discrimination within this genus (Bonvicino and Almeida 2000, Salazar-Bravo et al. 2013, Almeida et al. 2007, Bonvicino et al. 2010).
In Brazil, C. tener and C. expulsus are both widely distributed (Salazar-Bravo 2015).The first occurs mainly in the Cerrado and Atlantic Forest borders; and the second occurs in the Caatinga and Cerrado (Bonvicino et al. 2008).Sympatry between these species is common, especially in the central region of Brazil (Bonvicino et al. 2010, Salazar-Bravo 2015).The morphometry, karyotype, distribution, and ecological differences between these two species had been described by Bonvicino and Almeida (2000), but there is little data on the dental morphology of either species.These species have the same diploid number, but differ in the fundamental number (Bonvicino and Almeida 2000).Morphologically, these two species can be distinguished by their size, and some cranial measurements such as the lengths of the skull and molar series (Bonvicino et al. 2010), but owing to the size variation and ontogenetic development, species identification might be inaccurate (Hingst-Zaher et al. 2000).
Although karyotypes and gene sequences may be useful for discriminating among Calomys species (Bonvicino and Almeida 2000, Almeida et al. 2007, Bonvicino et al. 2010), these data are rarely available from fossil and subfossil material.Even considering that C. tener is a little smaller than C. expulsus in some cranial characters, is difficult to separate both species when the material available for study is incomplete.Fragmented material is often found in owl pellets, fossils and sub fossils, and Calomys remains are very common in these samples throughout South America (e.g., Pardiñas et al. 2000, 2002, Salles et al. 2006, Scheibler and Christoff 2007).
Calomys and other members of the tribe Phyllotini generally share simplified molars and complete loss of the mesoloph and mesolophid (Hershkovitz 1962), but discrete dental characters useful for identification of C. tener and C. expulsus are lacking (see Hershkovitz 1962).In the absence of such features and in face of the difficulties to identify either species with fragmentary remains, the variation between them need to be studied and verified using alternative quantitative tools.Many studies, not only involving rodents, have explored the use of morphometric methods with superposition of forms for iden-tifying different taxa, including molar analysis (e.g., Rohlf and Slice 1990, Bookstein 1991, Rohlf 1999, Becerra and Valdecasas 2004, Macholán 2006, Marcolini et al. 2009, Matthews and Stynder 2011).With this methodology, a complete identification is not always possible, but at least one can reduce the number of steps and time necessary for a correct identification (Becerra and Valdecasas 2004).In the Neotropics, however, studies investigating the usefulness of geometric morphometrics of the molar for taxonomic identification are unusual, and only skull and post-cranial elements are generally employed (e.g., Corti et al. 2001, Cordeiro-Estrela et al. 2006, 2008, Morgan 2009, Astúa et al. 2015).
The aim of present study is to explore geometric morphometric analysis to identify fragmented materials (modern and fossil) of small vertebrates based on molars; and to assess whether this method allows for accurate identifications.For this we applied this technique in one area: species-level identification of Calomys (C.expulsus and C. tener) based on their upper molars.

MATERIAL AND METHODS
The Calomys specimens analyzed were from owl pellets collected in the Natural Monument Peter Lund, Cordisburgo, in the central karst region of Minas Gerais, the Bambuí group, Brazil (Fig. 1).At the Natural Monument Peter Lund the pellets were collected inside the Salitre cave (19°07'17"S, 44°28'24"W) during the Park Management Plan.
In the Köppen climate classification system, the regional climate is Aw tropical humid, characterized by hot, rainy summers and dry winters (Travassos 2010).The average annual temperature in Cordisburgo is 22°C and the average annual precipitation ranges from 1250 to 1500 mm (Travassos 2010).In the karst area of Cordisburgo, much of the Cerrado has been replaced by agriculture and silviculture of Eucalyptus spp.(Travassos 2010).In the surrounding area and limestone outcrops, there is a semi-deciduous forest conditioned by the type of rock and climate, and it is possible to identify riparian and gallery forests along major drainages (Travassos 2010).
For species identification, we analyzed complete and fragmented skulls, maxillae and mandibles, examining the cranial sutures and morphology of the teeth.However, some fragments did not allow a consistent identification to the species level, thus restricting classification to the level of genus.The nomenclature of the species was based on that described by Patton et al. (2015).The molar teeth nomenclature was based on Reig (1977).All collected materials were deposited in the Mastozoology collection of the Museu de Zoologia João Moojen, Universidade Federal de Viçosa,Brazil (MZUFV 3861,MZUFV 3862,MZUFV 3863).
With the fragmented skull, the best characteristics used to differentiate C. tener from C. expulsus was the presence of alisphenoid strut, which is observed only in C. tener (Salazar-Bravo 2015) (Suppl.material 1), and the length of the molar series, which is on average 3.4 mm (range 3.1-3.9mm) in C. tener and on average 4.0 mm (range 3.8-4.2mm) in C. expulsus (Bonvicino andAlmeida 2000, Bonvicino et al. 2010).Due to an overlap in the measurements of the length of the molar series and difficulties to observe the presence of alisphenoid strut in all skulls due to fragmentation of the posterior part of the skull, some specimens were only identified as Calomys sp.
A total of 124 skulls were analyzed and separated into classes of dental wear, including 32 of C. tener, 57 of C. expulsus, and 35 of Calomys sp.(Suppl.material 2).For all analyses only the first right molar of Calomys skulls was analyzed, all of which had been previously identified as being of the same age based on wear class (see Suppl.materials 2).
The wear category class 2 showed the most numerically balanced sample between C. tener, C. expulsus, and Calomys sp.(Table 1) and, for that reason, was chosen for the morphometric analysis (see Suppl.material 3 for further information about the specimens analyzed).This subset included 46 specimens (15 of C. tener, 15 of C. expulsus and 16 of.Calomys sp.)Comparison of specimens from the same age category eliminated this potential source of age-related variation in enamel morphology due to tooth wear caused by chewing.Moreover, it minimized the allometric effect in the data, since the size of the specimens showed little variation.
The landmarks were digitalized using TpsDig v.2.17 (Rohlf 2015).We selected 17 landmarks corresponding to type II (points of maximum curvature, sensu Bookstein 1991), spanning the enamel folds connecting the main cusps of the first molar, which are described as follows (Fig. 2).Landmark 1, anterior extremity of anterolabial conule; Landmark 2, posterior extremity of anteroflexus; Landmark 3, anterior extremity of anterolabial conule; Landmark 4, lateral extremity of anterolabial conule; The multivariate analyses were performed using the Mor-phoJ v. 1.05f (Klingenberg 2011).After digitizing all landmarks using TpsDig, the centroid sizes of all specimens were calculated from the original coordinate matrix.The centroid size is the square root of the summed squared distances of each landmark to the centroid and can be used as a measure of general size of specimens (Zelditch et al. 2004).The matrices of coordinates were then superimposed using Procrustes standardization (the generalized least square method, GLS) to remove differences in size, orientation, and position between specimens (Zelditch et al. 2004).All further statistics were performed using the Procrustes residuals to analyze differences in shape and the centroid size values to evaluate differences in size.
For increased reliability of data, an analysis of error in digitization of anatomical landmarks was performed following the protocol of Adriaens (2007) using the TpsSmall v. 1.26 (Rohlf 2015).For this analysis, 12 specimens that were duplicated to evaluate the accuracy in digitization of anatomical landmarks were used.The results of the analysis of error are presented in Suppl.material 4.
From this experimental design, a preliminary principal component analysis (PCA) was performed using only species anatomically distinguishable, namely, 15 specimens of C. tener and 15 specimens of C. expulsus.This PCA was conducted to visualize the projection of individuals from the two axes of greatest variation, thereby detecting the distribution pattern in the graph.Then, we included the 16 specimens that could not be definitively identified as C. tener or may be C. expulsus.This addition aimed to increase the sample, totaling to 46 specimens, and explore the projection of Calomys sp. in the two axes of the PCA in the presence of distinguishable specimens.The second goal of adding specimens of Calomys sp. was to classify the unidentified specimens to one or another expected species (C.tener or C. expulsus).The wireframe, which helps in comparing the shapes of the specimens plotted in the PCA, was used whenever necessary to describe and discuss the shape of the specimens.
After each Calomys spp. was assigned to C. tener or C. expulsus through the pattern observed in the PCA, a Discriminant analysis (DA) was performed using MorphoJ v. 1.05f (Klingenberg 2011) and with 10.000 replications in the permutation test.As the Procrustes distances did not meet assumptions of normality, homoscedasticity and homogeneity of covariance (Suppl.material 5), the differences between the molar morphology of C. tener and C. expulsus were tested by PERMANOVA with adjustment of Euclidean distance (Anderson 2005).This analysis is analogous to ANOVA, however is a non-parametric test used to perform a comparison between two or more groups.In this test the significance is computed by permutation of group membership and p < 0.05 as the criterion of significance."Calomys sp." potentially identified as C. tener or C. expulsus into the PC1 vs. PC2 individual projections were also tested by PERMANOVA.Both analyzes were performed using 10.000 replications in permutation test using PAST v.3.11(Hammer et al. 2001).
Lastly, a pooled regression within each species (multivariate regression) was performed to evaluate whether the differences in the size of the tooth were correlated to the pattern of differentiation of the PCA projection.For this analysis, we used the centroid size of specimens as the independent variable and Procrustes coordinates as the dependent variable.Also, we performed a permutation test with 10,000 rounds to evaluate whether the dependence of shape (Procrustes distances) on the size (centroid size) of the tooth is significant.

RESULTS
The first PCA was performed only with C. expulsus and C. tener together.The PC1 and PC2 totaled 29.5% and 13.8% of the total variance, respectively, explaining 43.3% of the shape variation within the sample.It was possible to visualize either species separated in PC1 vs PC2 individual projections (Suppl.material 6).PERMANOVA test showed significant difference between C. expulsus and C. tener (p-value = 0.001 F = 6.079).The second PCA, which includes also the Calomys sp.specimens, showed a similar pattern as that of the previous PCA, with PC1 and PC2 holding respectively 28.7% and 12.3%, amounting 41% of the variation (Fig. 3).Discriminant analysis of first upper molar Procrustes coordinates classified 100% of the specimens into the correct groups; when the cross-validation technique was applied, 82.6% of specimens were recovered into the correct groups (Table 2).The permutation test using the Procrustes distance values (0.0484, p-value = < 0.0001) and Mahalanobis distances (6.4366, T-square = 0.0008, p-value = 0.0006) corroborated a distinction in the shape of the molar between species.The result of the PERMANOVA test for Calomys sp.(identified as C. tener or C. expulsus by PCA) together with the other specimens of C. expulsus and C. tener was significant (p-value = 0.0001 F = 8.283).
The variation expressed by PC1 showed differences in shape of the occlusal between both species.These differences are illustrated based on the wireframe view of mean shape (Fig. 4) and disparate shapes (Fig. 5 CT 02 (C.tener) can be distinguished from the other individuals by observing the shape differences explained by PC2 and illustrated in Fig. 6.CT 02 differed from the mean value for PC2 scores owing to the following morphological disparities: expansion of the anterior-posterior axis and slight retraction of the labial-medial axis of the anterolabial conule (landmarks 3 and 4); expansion of anterior-posterior axis in the protoflexus (landmark 5) (the anterior extremity of hypoflexus (landmark 7) did not expand proportionally), which caused a narrowed protocone in CT 02; slight retraction of posterior extremity of hypocone; and lingual expansion of metacone (landmark 13) and posterior displacement of anterior extremity of metaflexus (landmark 14).
To evaluate whether tooth size influences the patterns of variation in oclusal surface, we performed a multivariate regression with the Procrustes coordinates residual (shape) on the centroid size (size measurement).This regression showed a weak relationship that accounted for just 5.3% of the shape variation predicted by the size (Fig. 7), pursuant the permutation test accepting the null hypothesis of independence between the variables (p-value = 0.2118).

DISCUSSION
Calomys tener and C. expulsus are grouped into different clades (Almeida et al. 2007), however, neither species has discrete craniodental characters for species diagnosis (Hershkovitz 1962, Cordeiro-Estrela et al. 2006), and commonly used morphometric measures also overlapped (see Bonvicino et al. 2010).Nevertheless, the analyses performed in the present study indicate that the presence of the alisphenoid strut can be used to identify either species.
The overlap areas in morphometric cranial analyses using orthogonal matrix transformation techniques as PCA, irrespective of whether traditional or geometric, have also been identified in other species of other genera and in Calomys (Cordeiro-Estrela et al. 2006, 2008, Astúa 2009, Bonvicino et al. 2010, Martínez and Di Cola 2011).Furthermore, the results obtained in the present study for M1 shape are in agreement with those reported in other studies that showed a divergence in the shape of Calomys skull, but with small areas of intersection (Cordeiro-Estrela et al. 2006, 2008).
The main criterion for distinguishing C. tener from C. expulsus by traditional morphometrics is the size of the body (Bonvicino and Almeida 2000).Calomys expulsus can be differentiated from C. tener by having a larger body, smaller ear, proportionally smaller tail (generally), more robust skull, smaller braincase, length of molar row close to 4.0 mm in C. expulsus, 3.0 mm in C. tener (Winge 1888, Bonvicino andAlmeida 2000).However, Cordeiro-Estrela et al. (2006) analyzed the skull of C. expulsus and C. tener using geometric morphometrics and noted that allometry is not the main source of variation between the two species, despite its contribution to their differences.The main morphological differences between these two species are related to the patterns of shape variation (Cordeiro-Estrela et al. 2006), as observed in the present study through molar analysis.
As mentioned earlier, C. expulsus and C. tener occur in sympatry, and were differentiated in the present study through the presence of alisphenoid strut and length of maxillary molars.We observed a high similarity between our analysis and previous identification based on both characteristics, with more than 80% accuracy in DA.However, our results indicate that tooth size characters are not useful to diagnose either species.Thus, we suggest that the use of size characters (e.g., Bonvicino andAlmeida 2000, Bonvicino et al. 2010) to distinguish C. tener and C. expulsus should be avoided.Cordeiro-Estrela et al. (2008) arrived at a similar conclusion considering the skull centroid size of C. laucha and Calomys musculinus (Thomas, 1913).
In our results, CT02 deserves mention.This specimen was identified as C. tener because it had alisphenoid strut and a 3.7 mm long molar.However, the shape of this specimen, at least observing the differences explained by PC1, reveals more similarity with C. expulsus.Since CT 02 is a fragmentary specimen, our conclusion regarding its identification is limited.This specimen could be C. tener with larger molars or perhaps C. expulsus with alisphenoid strut and small molars.On the other hand, another possibility is that CT02 is a C. cerqueirai, a species described from near the investigation area of the present study (see Bonvicino et  al. 2010).However, we are not able to confirm this hypothesis, since C. cerqueirai shows no discrete cranial or dental characters or difference in size, when compared with C. expulsus (Bonvicino et al. 2010).As described by Bonvicino et al. (2010), the main differences between C. cerqueirai to C. expulsus are a Cinnamon-Brown overall dorsal coloration, head paler than dorsum, small ears with short brownish hairs, and sharply bicolored tail.Also, some descriptive cranial measurements, as palatal bridge, breath of incisive foramen, breadth of first maxillary molar, are overlapped.Unfortunately, we could not add C. cerqueirai specimens in our analysis and there are only twelve specimens deposited in other institutions (Bonvicino et al. 2010 Different molar morphologies in rodents can represent adaptations to different diets and can be relevant to avoid intraspecific competition (Parra et al. 1999, Renaud et al. 1999).This could mean that the differences in the shape of the molar, observed between C. tener and C. expulsus, observed in this study, could reflect different diets.This hypothesis needs testing, since the diet of C. expulsus and C. tener has not been published in the literature.A study on the outline of the first molar of extinct murine rodents demonstrated that different diet groups could be distinguished through geometric morphometric analysis (Cano et al. 2013).However, in marmots and some Didelphidae it has been demonstrated that diet is weakly correlated with molar shape (Caumul andPolly 2005, Chemisquy et al. 2015).
The possibility to perform taxonomic identifications based on molar shape, principally in rodents, is a very useful, especially when working with fossil materials (Macholán 2006, Barrón-Ortiz et al. 2008, Marcolini et al. 2009, Matthews and Stynder 2011).Matthews and Stynder (2011) distinguished Aethomys fossil specimens from extant specimens based on molar shape using geometric morphometric analysis; however, they emphasized that although the shape of the molar is important, a small sample size might hamper a complete analysis.Limitations in analyses owing to insufficient sample size are common, particularly in paleontological samples (Matthews and Stynder 2011), as it was in ours.Is important to point out that geometric morphometrics and features from the molars can be used in other types of studies beyond taxonomic identifications, as for example: correlating changes in molar shape with paleoclimate aspects (McGuire 2010), phylogenetic (Caumul and Polly 2005), morphological evolution and environmental variations (Renaud and Van Dam 2002).
Many studies using small mammal fossils or owl pellets for taxonomic identification are restricted to the level of genus (e.g.Salles et al. 1999, 2006, Bonvicino and Bezerra 2003, Scheibler and Christoff 2007, Rocha et al. 2011).With a good taxonomic identification it is possible to go beyond listing the species that occur at an area.The possibility to identify unknown specimens with an accurate morphology-based taxonomic assignments permits us to gain a better understanding of geographic distribution, migration and environmental response (Polly and Head 2004).Approaches using different quantitative, statistics and morphological analyzes can provide more scientifically accurate and reliable results (Polly and Head 2004).This is the first study using molars found in owl pellets for taxonomic identification through geometric morphometrics in South America.Our results were effective for the identification of dubious Calomys sp.specimens.All of them were identified either as C. tener or C. expulsus based on the individual projections of the PC1.The use of size as a diagnostic character is not effective for molar differentiation of C. tener and C. expulsus, as the regression test showed no evidence of correlation between size and shape.These results demonstrated the huge potential of geometric morphometric analysis as a tool in the rodent taxonomic identification using molars, especially in works with fragmented material (modern or fossil).This technique can be extensively useful principally by paleontologists, who have to identify isolated morphological elements, as molars of small vertebrates.This method can be also applied to other genera (e.g.Oligoryzomys Bangs, 1900 and Akodon Meyen, 1833) and to other groups such as bats and marsupials, wherever accurate taxonomic identifications of morphologically similarity and sympatric species are needed.

Figure 1 .
Figure 1.Map of Minas Gerais (Brazil), with the study location, municipality of Cordisburgo.
) of C. tener and C. expulsus.In general, C. tener showed a more retracted shape in the lingual-labial axis, due to an expansion of the cusps in C. expulsus compared to C. tener; flexus retraction and smaller aperture angles of the lingual flexus in C. expulsus compared to C. tener.The anterolingual conule of C. tener was retracted in the anterior-posterior axis compared to C. expulsus, was located more anteriorly, and was more oblique in relation to the anteromedian flexus.When compared with C. expulsus, C. tener had a retraction of the anterolabial conule in the labial-lingual axis and suffered another retraction in relation to the anterior and labial extremities (between the anatomical landmarks 3 and 4).The paraflexus of C. tener compared to C. expulsus was expanded in the anterior-posterior axis, and the paracone was retracted in the labial-lingual axis.The protoflexus of C. tener, when compared with that of C. expulsus, was expanded and the protocone was retracted on both the orientation axes.The metacone of C. tener, when compared with that of C. expulsus, was retracted in both the axes.The hypoflexus of C. tener, in relation to C. expulsus, was expanded and the hypocone was retracted in its anterior part.The posterior region of the molar of C. tener, in relation to C. expulsus, was retracted in the lingual region (landmarks 10, 11, and 12) and anterior-posterior axis.

Figures 4
Figures 4-6.Shape differences wireframes in PC1.Gray solid lines indicate C. tener (positive PC1 values) and black dotted lines indicate C. expulsus (negative PC1 values), see Fig. 3 for individual projections.(4) Differences between the mean shapes of C. tener and C. expulsus; (5) differences between the disparate shape of C. tener represented by CT 05 and C. expulsus represented by CE 09; (6) shape differences wireframes in PC2.Gray solid lines indicate the mean of shape of PC2 value and black dotted lines indicate specimen CT 02 (C.tener outlier).
, Mesquita and Passamani 2012, Colombi and Fagundes 2015.)But since neither species (C.cerqueirai and C. expulsus) has alisphenoid strut (Salazar-Bravo 2015), the most likely identification of specimen CT02 is outliner C. tener.We observed a pattern of differentiation in the oclusal shape of class 2, among sympatric specimens of C. expulsus and C. tener.In this class, the cusps of C. tener (as protocone and hipocone) were more retracted, whereas the cusps of C. expulsus are expanded in the lingual-labial axis (when compared with C. tener); the flexus (as paraflexus and protoflexus) are expanded and the aperture angles are larger in C. tener than in C. expulsus.

Table 1 .
Specimens examined and the respective sample sizes by dental wear category.