Shape Variation in Neotropical Cytheridella (Ostracoda) Using Semilandmarks-Based Geometric Morphometrics: A Methodological Approach and Possible Biogeographical Implications

Geometric morphometrics offer effective methods to obtain information about shape and shape variability. In ostracodology, landmark-based methods are, however, not well established. To test the applicability of geometric morphometric analyses for ostracods, we investigated shape variation among recent and fossil populations of the genus Cytheridella using a combination of landmarks and semilandmarks. The study focuses on the species' intraspecific morphological variability on a supra-regional scale, comparing living populations from Florida, Yucatán, Colombia and Brazil. We performed Generalized least-squares Procrustes Analysis on 508 adult and juvenile specimens (valves) including stages A-1 to A-4. The analyses show that the primary pattern in shape variation is ontogenetic allometry, supporting a clear separation of adults and juveniles. Shape changes are relatively small during ontogeny from A-4 to A-1. Greatest modification of valve shape occurs during the last molt phase. Insufficient differentiation of sexes is caused by females with less developed brood pouches. Disentangling size- and non-size-dependent shape changes reveals regional differences between populations of the species C. ilosvayi and supports its taxonomic distinction from a fossil relative (C. danielopoli). The distribution of regional morphotypes of C. ilosvayi in Florida, Mexico, and Brazil are congruent with fly ways of water birds.


Introduction
Morphometry is the quantitative description, analysis, and interpretation of shape and shape variation in biology [1][2][3]. Superior to the use of traditional measurements, geometric morphometric analyses purely address shape, i.e., the properties of an object that are invariant to scale, orientation and spatial position [3]. Two main approaches are available, one focusing on point-data, one on outlines and surfaces. Geometric morphometric analyses approach shape

Material
Subject of the study were the Cytheridella species C. ilosvayi Daday, 1905 and C. danielopoli Purper, 1979. Recent C. ilosvayi (type species of Cytheridella) have been sampled in several environments on public land in Florida, Mexico and Brazil (Fig 1) during field campaigns during 2009 to 2015. No Cytheridella species is listed in the IUCN Red List of Threatened Species (http://www.iucnredlist.org/) [39]. No permits were required for the described study, which complied with all relevant regulations.
In order to study similarities between extant Cytheridella and fossil relatives, we included C. danielopoli into the analyses. The fossil valves of C. danielopoli derive from late Miocene freshwater deposits at Eirunepé, SW Amazonas state, Brazil. Studied specimens were obtained by W. E. Piller and Martin Gross from the two outcrops Torre da Lua and Aquidaba (see [35,40]. All specimens are housed in the collection of the Museu Paraense Emílio Goeldi, Belém (Inv. No. MPEG-90-M to MPEG-193-M), additional material is stored at the Universalmuseum Joanneum, Department for Geology & Palaeontology, Graz (Inv. No. UMJG&P 210.903). All necessary permits were obtained for the described study, which complied with all relevant regulations. Permits for the fieldwork were issued by Conselho Nacional de Desenvolvimento Científico e Tecnológico/Ministério da Ciência e Tecnologia (CNPq/MCT; process number EXC 010389/2009-1).
Information on sampling localities is summarized in Table 1.
In order to exploit as much morphological information as possible, we used left and right valves from adults and instars, including stages A-1 to A-4 ( Table 2). In total, 508 valves were used for the analyses (247 right valves [RV], 261 left valves [LV]). Samples from Brazil (BB 01

Generalized Procrustes Analysis and Multivariate Statistics
Eight points were chosen as landmarks (LM; Fig 2). LMs 1-5 are type-I landmarks and characterize anterior pore tubuli; LM 6 is a type-II landmark defined by the dorsal dip point of the posterior curvature; LMs 7-8 are type-III landmarks delimitating the maximum anterior and posterior curvature. The valve outline was defined by two curves between LMs 7-8, each Table 1. Sampling localities with information about sample codes, locality names, country, sampling date and coordinates. The fossil C. danielopoli derived from fossil sites are referred to outcrops Aquidaba (AQ) and Torre da Lua (TO). For details see [35,40]. comprising 30 equidistantly spaced semilandmarks (Fig 2). Both landmarks and semilandmarks were set on digitized images with the program TpsDig v. 2.17 [41]. Right and left valves were investigated separately due to dimorphism in size and shape (adont hinge, left valves more bulgy; [36]). The sliders file to determine the sliding direction of the semilandmarks was created in TpsUtil v. 1.58 [42]. A generalized least-squares Procrustes Analysis was performed in TpsRelw v. 1.53 [43]. The analysis computes the consensus configuration, the partial warps, the relative warps, and centroid size. The semilandmarks are allowed to slide along the curves in order to minimize the bending energy between each specimen and the Procrustes average shape [11,44]. The deviations of the (semi)landmark configurations from the consensus were visualized by thin-plate spline deformation grids. Principal Components Analysis (PCA), i.e. the Relative Warps Analysis, was used to explore the morphospaces and Multivariate Analysis of Variance (MANOVA) and Canonical Variates Analysis (CVA) to test for significant differences between pre-defined groups (ontogeny/sex/ species). CVA fundamentally differs from PCA in that it requires specimens to be assigned to pre-defined groups, and then tests how well the data (e.g., warp scores) can be used to support those assignments. It aims at maximizing the ratio of the between-group variance to the within-group variance. Axes are scaled according to patterns of within-group variation and are not simple rotations of the original coordinate system as in PCA. Consequently, the distance between specimens or groups in a CVA-morphospace is not necessarily equal to their distance in the original morphospace (for details on the method see [4,45]). Since CVA is easily biased if the number of variables exceeds the number of specimens per group [45], we used for the computation the 15 first relative warps, which account for c. 97% of the shape variation in both valves. Due to low individual numbers C. danielopoli males were excluded. P-values of pairwise comparisons in MANOVA were Bonferroni-corrected.

Sample
In addition, we tested for morphological differences between populations from different geographic regions (and stratigraphic ages). To achieve a reasonable and statistically reliable comparison, we chose only female adult specimens for each valve, offering the highest number of specimens across all ontogenetic stages. The CVA was based on the 20 first relative warps resulting from the Generalized least-squares Procrustes Analysis.

Shape-Size Relationships
The geometric morphometric analysis disregards individual size. However, because of their allometric growth [47], morphological variability among the ostracods is to some extent expected to reflect size differences, especially with respect to differences between ontogenetic stages. To test for such a potential shape-size relationship, we performed ordinary least-squares regression analyses of valve (log 10 transformed) centroid sizes onto the first three relative warps for each valve in PAST.

Relative Warps Analysis
The relative warps analysis reveals, despite minor overlap, a clear distinction between adult and juvenile specimens in both left and right valves. Shape variations explained by the first three relative warps (RW) are 43.9%, 12.4% and 10.7% for left valves and 41.4%, 15% and 12.8% for right valves. Thin-plate spline grids for the minimum and maximum values and the outlier between axes of each of the first three RWs are displayed in Fig 3. Shape variation along RW 1 is primarily related to outline shape and the relative positions of the pore conuli (LMs 1-5). Positive scores correspond to enlarged posterodorsal shapes and a narrow arrangement of pore conuli. Along RW 2, positive values reflect more posterodorsally rounded and slightly convex ventral outlines. Negative values are associated with a decrease in the curvature of the posterodorsal shape as well as slight concave bend of the ventral outline. In left valves, RW 2 primarily differentiates between elongated (negative scores) and shortened valves (positive scores). LMs 1 to 6 play a minor role. In right valves, RW 2 parallels variation of the general shape (elongated with anterior inflation versus shortened with posterior inflation) as well as the relative positions of LMs 2-4 (pore conuli 2 to 4). In particular, LM 2 and 3 are closer in elongated valves, and more distant from each other in shortened valves.
Shape variation associated with RW 3 is also slightly different for left and right valves. Variation of left valves ranges from a convex (positive scores) to concave (negative scores) ventral shape. Moreover, the distance between pore conuli 2-4 varies. Especially LM 2 varies between a downwards shifted position and, thus, increasing distance to LM 3 (negative scores) and a closer position to LM 3 (positive scores). For right valves, in contrast, RW 3 reflects outline variation between shortened (positive scores) and elongated (negative scores) shapes. In addition, the position of the dorsal dip point of the posterior curvature (LM 6) varies between proximal (positive) and marginal (negative).
The distributions of the specimens show clear patterns only along RW 1 with respect to ontogeny. Adult individuals are largely separated from juveniles occupying higher positive (left valves) and negative scores (right valves), respectively. Juveniles are not clearly sorted according to their moult stage. Females and males overlap. Along RW 2, shape variation of females reflects different outline shapes ranging from forms with a well-developed brood pouch (positive scores, left and right valves) to elongated and less-developed brood pouches (negative scores, left and right valves). Males are characterized by negative scores reflecting their less curved posterodorsal shapes.
Morphological differences between C. ilosvayi and C. danielopoli are not discernible along RW 1, and their expression differs between left and right valves. The best differentiation between both species is achieved along RW 3 in left valves and along RW 2 in right valves. In both cases, the degree of separation increases with decreasing ontogenetic age: the two species are almost perfectly separated during stages A-3/4 and A-2, while they overlap in stages A-2/ A-1 and adult specimens. Only in right valves, males of both taxa seem to differ, but the low number of male C. danielopoli (n = 2) does not allow a reliable conclusion.
In summary, shape variation along RW 1 mainly corresponds to ontogenetic differences, while the differences between the two species are represented by RW 2 (right valves) and RW 3 (left valves).

Canonical Variates Analysis
The CVA based on the first fifteen relative warps of the whole dataset achieved partial delimitations of ontogenetic stages for both valves. The overall MANOVA statistics confirmed that the group means are significantly different (LV: Wilks' lambda test: Λ Wilks = 0.0081, F = 19.19, p < 0.001, Pillai's trace test: Λ Pillai = 2.81, F = 11.94, p < 0.001; RV: Wilks' lambda test: Λ Wilks = 0.0051, F = 21.01, p < 0.001, Pillai's trace test: Λ Pillai = 2.86, F = 11.77, p < 0.001). Adult males and females each form well-defined clusters with little overlap (LV: p < 0.001; RV: p < 0.001) and are also distinct from juvenile stages. Species are clearly separated with significantly different group means (Fig 4). Among the juveniles, however, delimitation of stages differs between both valves and species.
While the earliest stage of C. danielopoli (A-3 plus A-4) forms a more or less distinct cluster in right valves, they cannot be significantly distinguished from stage A-2 and A-1 in left valves (p = 1, p = 0.20, respectively). Earliest juveniles of C. ilosvayi can be differentiated significantly from A-1 in both valves (LV: p = 0.002; RV: p < 0.001), but not from A-2 (LV: p = 0.53; RV: p = 1).
Stages A-1 and A-2 overlap in both valves of both species, but only in left valves of C. danielopoli the group means are not significantly different (p = 1). All other pairwise comparisons between group means yielded significant differences.

Geographic and Interspecific Shape Variation in Female Specimens
In addition to the analyses based on the total dataset, we tested for differences of the morphological variability among populations of different geographic regions as well as between the living C. ilosvayi and the fossil C. danielopoli.
Except for a single outlier in left valves, there is no overlap between the species (Fig 5). The outlier represents a specimen with a short posterior part. The analysis indicates a broader morphological variability for C. ilosvayi than for C. danielopoli.
The general allometric trend observed corresponds to an elongation and increase of the posterior part in relation to the anterior part. In other words, a transition occurs from juveniles with a relatively pointed posterior to the reverse shape of adults with a relatively pointed anterior. Additionally, the position of the sulcus is displaced from a posteromedian position in juveniles to a more anteromedian position in adult specimens. Thereby, the distance between the sulcus and the pore conuli is reduced. Also, the anterior pore conuli shift to a more narrow position in adult individuals.

Ontogenetic Patterns
The succession of instars and sexes along the first relative warp indicates that the primary pattern in shape variation of Cytheridella species is the ontogenetic allometry. In general, shape change during the ontogeny of Cytheridella can be described as following: the anterior part of the valves is rounded in the earliest ontogenetic stage observed in this dataset, whereas the posterior part starts off more pointed (cf. Fig 3). Through ontogeny, the posterior part, more precisely the posterodorsal quadrant, enlarges. This development is traceable by LM 6, which represents the position of the transversal sulcus. Accordingly, there is a clear separation into adults and juveniles. Although the regression analyses showed that RW 1 is clearly tied to size, juveniles are not aligned according to instars along RW 1. This is also shown by the CVA (Fig 4), which illustrates similar patterns as the relative warps: adults and juveniles are separated; A-1 and A-2 overlap and A-3/4 build a distinct cluster only in right valves. In left valves all juvenile instars overlap.
This result indicates little shape change during the transition from A-4 to A-1. A comparably large morphological alteration happens during the last molt phase from A-1 to adults (Fig 4). Already previous studies have shown that the degree of shape change varies during the ontogeny of some species. The marine species Tyrrhenocythere amnicola (Sars, 1887) evinces greatest morphological changes between A-6 and A-5 [28]. Bradleya normani (Brady, 1866), in turn, shows greatest change between instars A-4 to A-3 and A-3 to A-2 [48]. Our dataset comprises juveniles only down to A-4; it is, thus, possible that distinct shape changes, which might have occurred earlier during ontogeny, remained unnoticed. In comparison, ontogenetic shape changes are similar for the two species C. danielopoli and C. ilosvayi, confirming their close relationship.
Another feature of the juveniles is their relatively high variability compared to adult specimens, and it is observable along each of the three first relative warps (Fig 3). Under environmentally controlled conditions and genetic homogeneity the cypridoidean species E. virens exhibited a similarly high individual shape variation of juveniles which decreased from A-4 onwards. It has been suggested that environmental factors play a major role in shape determination during ontogeny [49].
The insufficient differentiation of sexes in C. ilosvayi is caused by females with a less developed brood pouch, resembling the morphology of males. This result undoubtedly has important consequences in sex identification in fossil samples and indicates the need to assess traits involved in sexual dimorphism before making inferences [33]. Morphological differentiation between sexes is achieved through males which have concave ventral shapes while females display a slight depression dorsally and both, convex and concave ventral outlines. Contrasting to C. ilosvayi, sexes are not differentiated properly in C. danielopol probably due to relatively large and elongated posterior shapes of males and resembling, thus, females.

Species Differentiation
The CVA in Fig 4 and associated MANOVA demonstrate clearly the offset between the two species compared to their intraspecific variability. Except for single outliers in left valves the two species do not overlap. In summary, C. danielopoli is characterized by more elongated shapes and straight ventral shapes. Specific differences between juveniles of the two species are observed from the relative warps analysis, where they are almost completely separated. Juveniles of C. danielopoli are also characterized by straight ventral outlines with an obtuse posterior compared to the more pointed posterior of C. ilosvayi. More juvenile specimens are needed to test at which instars species differentiation is significantly achieved.

Geographic Shape Variability
Apart from the clear specific differences, the CVA of females also indicates geographic shape variability in C. ilosvayi (Fig 5). Interestingly, the morphological differences are higher between populations from Mexico and Florida than between those from Brazil and Florida. Although the groups have significantly different group means, the CVA indicates widely overlapping clusters. Hence, the analyses indicate the presence of distinct regional morphotypes, but a clear differentiation is not achieved. Even at the population level (female) morphotypes were found. These morphotypes are primarily discriminated by valves size but reveal also distinct differences in their soft part morphologies (i.e., podomere ratios of antennula and antenna [38]).
Further studies including males and juveniles are required to explore regional differences and allow a taxonomic classification of these morphotypes. This may ultimately help to identify biogeographical and/or phylogeographic patterns in Cytheridella.
For now, we can only speculate about the causes for the present disjunct distribution of C. ilosvayi because information about its origin, dispersal and fossil record are limited. Given the patchy nature of freshwater environments, long-distance dispersal in freshwater ostracods relies on passive dispersal. Several studies have stressed the importance of waterfowl-mediated dispersal of Ostracoda (e.g., [50][51][52][53]). Individuals are transported occasionally across long distances, internally or externally, often involving desiccation-resistant eggs but also living ostracods [51]. The ability to produce resting eggs is well documented from Cypridoidea but has been reported for some Limnocytheroidea, the family to which Cytheridella belongs, as well [54].
Accepting the preliminary hypothesis that C. ilosvayi has been dispersed via water birds, the patterns of similarities and differences of the regional morphotypes can be viewed in a different light. Instead of paralleling geographic distance, they might mirror the structure of present waterfowl flyways across the Americas (e.g., [55]). Four major routes are indicated for migrating water birds from north to south, i.e., the Pacific, Central, Mississippi and Atlantic flyways [56]. Although we lack information about the geographic origin of the species and the age of the populations, the present patterns suggest that dispersal of C. ilosvayi between Florida, Mexico and Brazil may have happened via the Mississippi or Atlantic flyways. Moreover, the disjunct distribution is likely a result of multiple independent dispersal phases. The fact that the morphological differences are higher between populations from Mexico and Florida than between those from Brazil and Florida argues for a longer geographic separation.

Conclusion
The outcome of the present investigation shows the potential of combining characteristics of the outline and surface to study ostracod valve morphology. So far, studies have focused either solely on the ostracod outline (e.g., [14,18,22,31,57]) or on specific features of the valve (e.g., [12,14,16,18,22,31,48,57]). The geometric morphometric approach applied here enabled the detailed documentation of the patterns of allometric growth in two species of Cytheridella, the discrimination of closely related species, as well as the differentiation of regional morphotypes in one of them. More precisely, our results indicate that ontogenetic shape variation in Cytheridella is largely owed to outline variation as well as the position of the transversal sulcus and the relative distance between the pore conuli. The patterns of allometric growth are similar in recent C. ilosvayi and fossil C. danielopoli: comparably small morphological change between juvenile instars is followed by a major morphological alteration during the last molt phase. While this analogy confirms the close relationship of both species, the analysis supports their discrimination based on elongated outline and straight ventral shape. Furthermore, regional shape variation is detected for C. ilosvayi, evidencing the presence of distinct morphotypes in populations from Florida, Mexico and Brazil. The populations are, however, not fully discriminated, and a taxonomic conclusion still requires the evaluation of additional data, including information from other instars as well.
The results clearly encourage the utilization of both valves in morphometric analysis, due to mutative development of shape variables in left and right valves. We hope that this study stimulates the use of geometric morphometric analyses in ostracodology. Unlike many other morphometric approaches, it supports the combination of outline and landmark features and, above all, it is an easy and quick method to address shape and shape variability.
Finally, the identification of regional morphotypes of C. ilovayi reveals a biogeographical pattern that can be interpreted to reflect water bird's flyways across the Americas.

Data Reporting
Data are available at the permanent web site of the Institute of Earth Sciences, University of Graz. Data include the.tps files for left and right valves and the respective results of Relative Warps Analysis and CVA. Files are readily accessible under the following URL: http:// iewarchiv.uni-graz.at/zusatz/NeotropicalCytheridella/.