Morphometric convergence among European sand gobies in freshwater (Gobiiformes: Gobionellidae)

Abstract The five genera of sand gobies inhabit the seas and freshwaters of Europe and western Asia and occupy habitats ranging from fully marine to exclusively freshwater. In this study, we use geometric morphometrics to quantify body shape among sand gobies, in order to investigate how shape has evolved and how it is related to habitat. We also compare body shape between preserved museum specimens and fresh specimens, to determine whether or not fixation and storage in ethanol introduce detectable bias. We confirm that the fixed specimens exhibit significant shape changes as compared to fresh specimens, and so, we perform the bulk of our analyses exclusively on fixed specimens. We find that Economidichthys, Orsinigobius, and Pomatoschistus occupy distinct regions of morphospace. Knipowitschia and Ninnigobius have intermediate forms that overlap with Pomatoschistus and Orsinigobius, but not Economidichthys. This pattern is also in rough accordance with their habitats: Pomatoschistus is fully marine, Economidichthys fully freshwater, and the others fresh with some brackish tolerance. We augment a recent phylogeny of sand gobies with data for P. quagga and interpret morphometric shape change on that tree. We then evaluate convergence in form among disparate lineages of freshwater species by constructing a phylomorphospace and applying pattern‐based (convevol) measures of convergence. We find that freshwater taxa occupy a mostly separate region of morphospace from marine taxa and exhibit significant convergence in form. Freshwater taxa are characterized by relatively larger heads and stockier bodies than their marine relatives, potentially due to a common pattern of heterochronic size reduction.

Apart from Ninnigobius, the remainder of the sand gobies fall into two groups, the marine Pomatoschistus species and a clade containing the more brackish to freshwater Knipowitschia, Orsinigobius, and Economidichthys. In this study, we explore the significance of overall body shape in the sand gobies. We use geometric morphometrics to quantify body shape and then analyze the distribution of shape changes among species, between habitats, and in the context of their phylogeny. We also use pattern-based tests of convergence to evaluate whether the freshwater species, which do not form a clade, exhibit significant convergence in body form. To provide an evolutionary framework for these tests, we generate a new sand goby phylogeny, adding to the dataset of Thacker et al. (2019) with newly generated sequence data for the enigmatic species P. quagga, derived from the study of Öztürk and Engin (2019).
Freshwater fish species face a different set of selection pressures from their habitats than marine fishes, given that their environment is usually smaller, more bounded, and potentially experiences more variable flow regimes. There are no overall generalizations concerning freshwater as opposed to marine fish body form, but among sand gobies there are some qualitative differences. Photographs of representative species from each sand goby genus are given in Figure 1. Economidichthys and Orsinigobius are deeper-bodied, with proportionally larger heads, than their counterparts in Pomatoschistus and Knipowitschia. They are also generally smaller, attaining body sizes of 30-50 mm as adults, compared to 60-110 mm in most Pomatoschistus species, with Knipowitschia and Ninnigobius intermediate at 30-70 mm. Common morphological traits among freshwater species may relate directly to their ecology or may simply be a consequence of the smaller body size they all attain. Smaller fish species, particularly if they have achieved the size reduction by heterochrony, would be expected to have proportionally larger heads as well as reductions in characters such as scalation and head pores (Weitzman & Vari, 1988). Such reductions are seen in freshwater sand gobies (Miller, 1990). Miniature species may also exhibit unusual morphological novelties, such as the perianal organ found uniquely in Economidichthys species (Economidis & Miller, 1990;Hanken & Wake, 1993). The discrete characters associated with freshwater habitat preference in sand gobies are known, but the overall shape changes among the species have not been investigated. We seek to quantify the overall body shape among sand goby species and then interpret those patterns in the context of a phylogeny. This evolutionary perspective enables us to determine whether there are common aspects of shape change among freshwater sand goby species, and evaluate whether and to what degree those changes are concordant with the phylogeny, or convergent across lineages.
Morphological similarity among living organisms may result from a variety of causes, but inference of mechanism requires first quantifying the similarity, then ruling out the possibility that similar traits are simply due to inheritance from a common ancestor that also had those traits (Sanderson & Hufford, 1996). If similarity is confirmed, and phylogenetic commonality is ruled out, then it is reasonable to postulate other explanations for the pattern, including evolutionary convergence. Convergence may arise from a common selective pressure, functional constraint, ontogenetic constraint, some combination of these, or simply by chance (Losos, 2011 (Rüber & Adams, 2001), and convergence of head and jaw shape among herbivorous lizards (Stayton, 2006). In those cases, similar morphologies are exhibited in multiple, phylogenetically independent lineages (convergent pattern), and because the traits in question are easily relatable to function, a reasonable link between pattern and evolutionary process (parallel response to a common selective pressure) may be hypothesized. In other cases, convergence may be the result of a common evolutionary response induced by some separate selective change, such as patterns of trait simplification resulting from paedomorphic size reduction (Losos, 2011). Among sand gobies, the marine and freshwater species exhibit few consistent differences, but the overall body shapes do vary qualitatively.
Using morphometric analysis coupled with phylogeny, we are able to evaluate the pattern of morphological evolution among these species and infer the evolutionary timing and cause.
Although many sand gobies are not uncommon where they occur, they are not often collected and so are generally rare in museum collections. We use ethanol-fixed and preserved specimens from museum collections in this study and additionally compare the morphometric patterns to those derived from fresh specimens. We use a photographic dataset of six sand goby species  (Risso, 1810)) from live collections made across the coastal localities and fresh waters of Greece and the Adriatic (Venice Lagoon). We compare those individuals to fixed specimens of the same species to determine whether or not fixation introduces a bias in analysis of landmark data, and if so, whether or not that bias is consistent and what form it takes. Previous works comparing geometric morphometric analysis of fixed and fresh individuals from both marine and freshwater fish species determined that fixation did introduce a significant bias, specifically manifested as overall shrinkage and decrease in eye diameter .
To evaluate evolutionary patterns in body form, we use a more comprehensive dataset of fourteen sand goby species, the data for which are all derived from fixed specimens. These species include the most common and widespread sand goby species and span all the clades within the sand goby phylogeny of Thacker et al. (2019). Due to the rarity of sand gobies in museum collections, and the need for intact adult specimens for morphometric analysis, we were unable to assemble data for the rarer species. We examined between three and 26 individuals for each species used, except for Pomatoschistus lozanoi (de Buen 1923) for which only one appropriate specimen was available. We then combine those data with a phylogenetic hypothesis, augmented from a previous sand goby phylogeny (Thacker et al., 2019). We evaluate the phylogenetic significance of shape change and investigate the degree to which freshwater species have converged.

| Landmark acquisition and analysis
We examined a total of 245 sand goby specimens from 14 species, 127 from museum collections (ethanol fixed and preserved) and 118 freshly caught, photographed shortly after death. Fixed specimens were examined at the Naturhistorisches Museum, Vienna, and the Natural History Museum, London, and additional specimen photographs were provided by the Muséum National d'Histoire Naturelle, Paris. The specimens were collected between 1874 and 1995, with most dating back to before the early 1900s, at localities spanning the northeastern Atlantic, Mediterranean, and Adriatic Seas and associated freshwaters. We selected specimens that were adult, undamaged, and as unbent as possible and photographed all specimens in left lateral view with an Olympus Tough TG-5 camera, using a copystand and the automatic z-stacking option. Although many of the We digitized 17 landmarks for each individual, using ImageJ version 1.52 a (Schneider, Rasband, & Eliceiri, 2012), as shown in Figure 2. This suite of external landmarks describes the overall body shape, the fin positioning, and the locations and size of the mouth and eyes. They have been used previously to quantify shape variation in gobioid species (Thacker, 2014(Thacker, , 2017 and are reliably assignable to both preserved and fresh specimens. All landmarks were assigned in all specimens examined. We then forwarded the landmark coordinates to MorphoJ version 1.05d (Klingenberg, 2011), performed a Procrustes fit, generated a covariance matrix, and used that matrix as input for principal components analysis (PCA).
We first analyzed paired landmark data for six species for which we had both preserved and fresh specimen data: Economidichthys pygmaeus, Knipowitschia caucasica, K. milleri, K. panizzae, Ninnigobius canestrinii, and Pomatoschistus marmoratus. For those data, we used PCA to evaluate whether or not individuals of the same species could be separated on the basis of preservation. We additionally used MorphoJ to perform a discriminant function analysis (DFA), grouping individuals by preservation technique, to compare morphometric distance between fixed and fresh specimens and assess its significance. For both the PCA and the DFA, we analyzed each of the six species separately and also all of the species together. Finally, we imported the landmark data into R (version 3.5.0) and used geomorph (version 3.0.7; Adams & Otarola-Castillo, 2013), to perform Procrustes ANOVA (ANOVA of Procrustes coordinates) and further test for significant shape divergence between the fixed and fresh individuals.
We found that preservation technique introduced a significant bias to the shape data, so we proceeded with our morphometric and comparative analyses using only fixed museum specimens.  Table 1, and museum catalog numbers for material examined are given in the Appendix (Table 3). For these data, we digitized landmarks and again used MorphoJ to generate a PCA, as well as a canonical variates analysis (CVA) grouping the individuals by genus and by habitat preference (fresh, brackish, or marine). We performed Procrustes ANOVA (not corrected for phylogenetic relationships) with geomorph, regressing shape change against species, genus, and habitat.  (Table 4). We assembled the matrix using Geneious (Biomatters, Ltd.) version 10.2.6 and performed a Bayesian search as originally described: 10 × 10 7 generations using a GTR + I + G substitution model, run with four simultaneous chains, sampling every 1,000 replications and discarding the first 10% of trees as burn-in.

| Phylogenetic comparative analyses and tests of convergence
We constructed a 50% majority-rule consensus tree, trimmed the hypothesis to two exemplars of each species (three for widespread P. marmoratus), and then calibrated the phylogeny using BEAST  (Garland, Dickerman, Janis, & Jones, 1993 For habitat and range information, we scored each species using information from field guides (Kottelat & Freyhof, 2007;Louisy, 2015;Miller, 2004;Miller & Loates, 1997) and FishBase (Froese & Pauly, 2018). We performed the habitat coding in two ways. First, we coded habitat as a multistate character, encompassing exclusively freshwater, freshwater/brackish, freshwater/brackish/marine, brackish/marine, or exclusively marine. This coding is more accurate, but introduces an artifact in that overlapping conditions (for instance, freshwater vs. freshwater/brackish) are treated as independent states. We used this multistate coding for the Procrustes ANOVA and phylogenetic MANOVA. We also performed the Procrustes ANOVA, phylogenetic MANOVA, and convergence tests with a binary coding scheme, using marine for the exclusively marine Pomatoschistus species, and freshwater for the other species, all of which inhabit fresh or brackish water. This binary scheme should more accurately test our primary question, which is whether or not species that are exclusively or mostly known from freshwater exhibit any common morphometric pattern.

| Morphometrics of fixed and fresh specimens
The morphometric PCA plots for fresh versus fixed specimens are given in Figure 3. In most of the comparisons (except  Figure 3 because it is much easier to interpret the graphical patterns if the separate species are not overlain. In both the combined and separate PCAs, the wireframes displaying shape change on PC1 and PC2 all describe changes in body depth and tail length and curvature, known artifacts of fixation on body shape Martinez et al., 2013). DFAs comparing fixed and fresh individuals, with species analyzed separately or together, were significantly different based on 1,000 permutations of the T-square statistic (p < 0.0001) for every species except for P. marmoratus (p = 0.098). Procrustes ANOVAs of shape change between fixed and fresh specimens of each species were nearly significant (p = 0.056, N. canestrinii), significant (p = 0.01, K. panizzae) or highly significant (p = 0.001-0.007, all other species, consistent with greater statistical power associated with larger sample sizes).
To further evaluate the effect of fixation on shape change, we examined the Procrustes distance between fixed and fresh specimens for each species. We first compared the Procrustes distances obtained from the separate and combined DFAs, and for each species, they were identical to the third decimal place. Distances between fixed and fresh specimens ranged from 0.034 (Economidichthys pygmaeus) to 0.067 (Knipowitschia caucasica), with a mean of 0.052.
These estimates are comparable to the distances among fixed species, which range from 0.026-0.132, with a mean of 0.065. Given these artifacts, we did not combine the fixed and fresh specimens into a single dataset for further analysis. Instead, we proceeded with exclusively fixed museum specimens.

| Sand goby phylogeny
The phylogeny of sand gobies is given in Figure 4. This hypothesis is based on the same data as Thacker et al. (2019) The calibrated phylogenetic hypothesis, based on representative individuals for each species, is given in Figure 5.

| Sand goby morphometrics
The dataset of fixed museum specimens included 14 species of sand gobies and encompassed most of the widespread common species found in the seas and freshwaters of Europe. Plots for PC1 versus PC2 of morphometric data are given in Figure 6. The first two PC axes account for 50% of the total variance (35% and 15%, respectively

| Phylogenetic comparative analyses and tests of convergence
We detected significant phylogenetic signal among the shape data  Table 2.
The phylomorphospace of sand goby species is shown in Figure 7.
Generally, the freshwater genera (Knipowitschia, Orsinigobius, and Economidichthys) occupy the right side of the phylomorphospace, with marine Pomatoschistus species arrayed on the left side.
The exception is Ninnigobius, which is placed slightly inside the Pomatoschistus morphospace, nearer to P. pictus and P. marmoratus than to the other freshwater genera. The separation between these groups is most strongly loaded on PC1, with PC2 representing elongation in the caudal region and providing differentiation among species within the two ecological types.

| Morphometric distinctions between fixed and fresh specimens
Sand gobies are small and elongate and so may be particularly affected by the shrinkage and deformation induced by fixation. Our comparisons of fresh and fixed individuals of six sand goby species confirm this pattern. Four out of the six species examined exhibit complete or near-complete separation between fixed and fresh specimens on plots of PC1 versus PC2 (Figure 3), both Procrustes ANOVAs and DFAs yielded significant or near-significant differences in every comparison, and Procrustes distances among fixed and fresh specimens of the same species were comparable to the distances between fixed species. Fixation in ethanol has been shown to yield a small but significant reduction in body length in both marine and freshwater fishes, with most effects seen in the first 20 days after preservation (Buchheister & Wilson, 2005;Fey & Hare, 2005;König & Borcherding, 2012;Thorstad et al., 2007).
Differences in overall shape, quantified with geometric morphometrics, were also confirmed for formalin fixation in two marine species  and for ethanol fixation in a freshwater species . Both of those studies found that significant errors were introduced into the shape data by fixation, most prominently manifested as a shrinkage in the eyes and overall body depth. In our more terete, elongate specimens, we confirm reductions in body depth and overall length for fixed specimens, as well as deformation of the caudal region. There are slight differences in relative eye placement between the fixed and fresh specimens, but we did not detect notable shrinkage of the eyes. Interestingly, the changes in shape due to fixation were not consistent across species. Four of the six fixed/fresh specimen comparisons displayed an increase in body flexion and a decrease in overall length in the fixed individuals (Knipowitschia caucasica, K. panizzae, Ninnigobius canestrinii, and Pomatoschistus marmoratus; Figure 3), but the other two (Economidichthys pygmaeus and K. milleri) had fixed and fresh individuals overlapping in the plots, without clear distinction between the groups. This may be due to the fact that PC eigenvectors, in contrast to simple linear measurements, describe complex overall shape changes and may vary in which axis of change they detect as the primary one. However, in all comparisons, the DFA's and Procrustes ANOVAs showed significant or near-significant differences between fixed and fresh specimens. To avoid any confounding effects of including both fixed and fresh specimens, and because the majority of specimens available for morphometric analyses were fixed museum specimens, we proceeded with fixed specimens only.

| Sand goby phylogeny and timing
Our Our hypothesis indicates that sand gobies arose in the late Oligocene (26.1 Mya), followed by a radiation of genera (and the P. quagga lineage) up to the mid-Miocene (the split between Knipowitschia and F I G U R E 5 Calibrated sand goby phylogeny, trimmed from that shown in Figure 4 to a total of 51 individuals, 45 sand gobies plus six outgroup sequences, and calibrated with three fossil and one legacy calibrations. Black circles at nodes indicate 95%-100% posterior probability, and error bars indicate 95% highest posterior density of calibration estimates. Among the sand gobies, blue shading on species names indicates marine habitat, green shading indicates freshwater tolerance (FW or FW/BR habitat preference), and species examined for morphometric shape data are circled. to the emergence of the Greek peninsula as sea levels fell (Rögl, 1999a(Rögl, , 1999b Flecker, Baak, Lunt, & Krijgsman, 2016;Rögl, 1998) (Marzocchi et al., 2016;Orszag-Sperber, 2006

| Evolutionary shape change and convergence among freshwater sand gobies
Sand gobies inhabiting fresh to brackish waters are not monophyletic, each genus is independently derived and they are interspersed phylogenetically with marine Pomatoschistus species, including a separate lineage for P. quagga, as shown in Figure 5. They also arose and then diversified at different times throughout the Miocene, Pliocene, and Pleistocene. Morphometrically, all of the freshwater species occupy a distinct region of morphospace, with the exception of Ninnigobius, the earliest-diverging genus. Ninnigobius has a more generalized morphology than the other freshwater species ( Figure 1); in particular, the relative enlargement of the head is less pronounced. CVA's and Procrustes ANOVAs of shape data (not phylogenetically corrected) show significant divergence among species in different habitats, but the distinction is not significant when corrected for phylogenetic relatedness. However, the pattern-based tests of convergence among the freshwater species were highly significant, with convergence accounting for 40.2% of the variation.
Unsurprisingly, variation in form among these species includes both phylogenetic and convergent components.
Convergence in morphology may be the result of a variety of evolutionary processes, including response to a common selective pressure, restriction by some developmental, genetic, or functional constraint, or even simple coincidence (Burns & Sidlauskas, 2019;Sanderson & Hufford, 1996 a smaller body size would be favored, for reasons including lower food requirements, ease of camouflage, and potentially more rapid growth and shorter generation time. In particular, if there is pressure for a more rapid maturation, which would be favored in a less stable habitat, it is possible that the small body size and proportional changes in the head and body shape are the result of heterochrony. A quickening of development in freshwater sand goby species could result in the overall convergent morphological changes documented here, even though the shape changes are not themselves the targets of selection. Freshwater sand gobies share reductive morphological changes in the head canals and scalation that are characteristic of subadult developmental stages (Economidis & Miller, 1990) and attain a generally smaller size. Lifespan data are scarce for these species, but Ninnigobius canestrinii and both Economidichthys species have lifespans of only one year, Knipowitschia caucasica may live for 1-2 years, and Pomatoschistus microps can live for 2-3 years (Miller, 1990(Miller, , 2004.
Overall, the sand goby genera exhibit a continuum of shorter lifespan, morphological reduction, and proportional enlargement of the head and shortening of the body that is correlated with increased freshwater preference, and consistent with a common heterochronic process of paedomorphosis.

ACK N OWLED G M ENTS
We are grateful to the curators and collections managers of museums that provided specimens and photographs of sand gobies:

CO N FLI C T O F I NTE R E S T
The authors declare no conflicts of interest.

AUTH O R S CO NTR I B UTI O N
CT examined specimens and collected morphometric data. CG performed fieldwork and collected DNA data. CT performed analyses.
Both authors wrote and edited the manuscript.

DATA AVA I L A B I L I T Y
Museum specimens: Specimens examined for this study are listed in the Appendix, 3A1. DNA sequences: GenBank accession numbers are listed in the Appendix, 4A2. DNA alignment and phylogeny, and morphometric data are available from Dryad https ://doi. org/10.5061/dryad.c0qt85m.