Identification of past and present gobies: distinguishing Gobius and Pomatoschistus (Teleostei: Gobioidei) species using characters of otoliths, meristics and body morphometry

Gobies (Gobiidae + Oxudercidae) are among the largest groups of extant marine fishes. Fossils of gobies are abundant since the Miocene, and many species have been reported so far. However, delimitation of fossil goby species is challenging because molecular markers and diagnostic traits such as the disposition of sensory head papillae are lost. This study provides, for the first time, an actualistic framework for the identification of fossil goby species. We focus on characters that can in principle be recognized in fossils, and evaluate their ability to discriminate between extant goby species based on statistical analyses. Using 14 extant species of Gobius and seven species of Pomatoschistus , we meristic counts. In addition, the otoliths of all species are depicted based on SEM images and briefly described. Otolith Fourier shape analysis proved to be most efficient in discrimination of species within both genera, Gobius and Pomatoschistus . Several characters used in the other approaches also worked well, but the results were variable, and the relative taxonomic significance of particular variables tended to change depending on the species under consideration. We propose otolith shape analysis as a powerful tool to explore ancient goby species diversity when samples with abundant fossil otoliths are present. Overall, the herein presented data will greatly facilitate delimitation of fossil goby species in future studies, and will consequently shed new light on the evolution of goby diversity and biogeography through time.


Introduction
Gobiidae and Oxudercidae ("gobies" hereafter) are generally small fishes (length <25 cm), and together they constitute one of the most species-rich groups within the modern bony fishes (Teleostei) (Nelson et al., 2016), with >1,250 species in the Gobiidae and >650 species in the Oxudercidae (Fricke et al., 2019). Gobies are globally distributed in diverse marine, brackish and freshwater habitats, and also play important roles in reef ecology (Patzner et al., 2011;Tornabene et al., 2015;Brandl et al., 2018). Many species are highly specialized including close associations with crustaceans (Karplus, 1987;Karplus et al., 2011; or as part of the cryptobenthic community (Glavičić et al., 2016;Brandl et al., 2018). The European Gobiidae and Oxudercidae comprise three distinct lineages, which were named Aphia-lineage, Gobius-lineage, and Pomatoschistus-lineage (Agorreta et al., 2013;Thacker, 2015). In the present work, we focus on the name giving genera of the latter two lineages, i.e., Gobius Linnaeus, 1758 and Pomatoschistus Gill, 1863, which are the most species-rich genera of European gobies (Miller, 1986;Patzner et al., 2011). Their lifestyle is usually benthic or cryptobenthic, with a preference for sandy and rocky substrates (Macpherson & Duarte, 1994;Miller, 2004;Patzner et al., 2011). The genus Gobius currently comprises 28 recognized species, which are mainly distributed in the NE Atlantic, the Mediterranean and the Black Sea (Miller, 1986;Patzner et al., 2011;Froese & Pauly, 2019). Notably, two of these species are recent discoveries (Iglésias et al., 2016;Kovačić & Šanda, 2016). Fourteen species are traditionally attributed to the genus Pomatoschistus, including three that were only recognized in the last decade (Miller & Šanda, 2008;Engin & Innal, 2017;Engin & Seyhan, 2017;Froese & Pauly, 2019); however, the genus is not monophyletic according to recent phylogenetic studies (Thacker et al., 2019). Species of Pomatoschistus mostly occur in coastal habitats along the NE Atlantic, the Baltic Sea, and in some areas of the Mediterranean and the Black Sea (Miller, 1986;Patzner et al., 2011). Identification of the individual species in both genera, Gobius and Pomatoschistus, relies on coloration patterns, the arrangement of the head sensory organs (papillae), and combinations of various other characteristics such as squamation, morphometric and Downloaded from Brill.com11/03/2020 07:41:22AM via free access meristic traits; species determination often turns out to be very difficult, especially between closely related species (Miller, 1974(Miller, , 1986(Miller, , 2004Kovačić & Šanda, 2016;Engin & Seyhan, 2017). Fossils are the only direct source of information on ancient species diversity and biogeography, and thus provide essential insights into the evolutionary history of taxa. Fossils of gobies are abundant since the early Miocene (23 Ma) and comprise both articulated skeletons (sometimes with otoliths in situ) and isolated otoliths (e.g., Weiler, 1963;Radwańska, 1992;Bratishko et al., 2015). Records of fossil otoliths are much more common than discoveries of fossil skeletons, and this explains why about two-thirds of the known fossil goby species have been described solely on the basis of otoliths (Gierl et al., 2018). But the delimitation of a fossil goby species is often problematic because information on the range of variation in skeletal and otolith traits within a respective species and between related taxa is restricted. Furthermore, the characters of the skeleton may vary little between extant goby species (Miller, 1986;Kovačić & Šanda, 2016). This, together with the loss of delicate characters in fossil skeletal material due to taphonomic processes, can lead to underestimation of the number of fossil species (see Gaudant & Quayle, 1988vs. Gierl & Reichenbacher, 2017Gaudant, 1998vs. Bradić-Milinović et al., 2019. Otoliths of gobies have the advantage that they are quite robust and do not experience major loss of informative characters during the fossilisation process. On the other hand, they usually display a high degree of morphological variability, which can make their taxonomic interpretation challenging (Malz, 1978;Nolf, 1985;Jost et al., 2015). In some cases, subtle differences in otolith morphology may allow the identification of several species (Schwarzhans, 2014; Gierl & Reichenbacher, 2015;Reichenbacher et al., 2019), while in others clear differences in otolith morphology may reflect variation in a single species living in different environments (Vasil'eva et al., 2016). The objective of this study is to provide, for the first time, an actualistic framework for the taxonomic value of characters that can in principle be recognized in fossil gobies, i.e., otoliths, body morphometry and meristics. We used 14 extant species of Gobius and 7 extant species of Pomatoschistus to test the suitability of our approach.

Samples and datasets
Fourteen species of Gobius and seven species of Pomatoschistus were collected from sites along the coasts of the Mediterranean and the Baltic Sea. Details of sites and numbers of individuals are provided in table 1, collection numbers are available in the supplementary table S1. Specimens were collected with hand nets during scuba diving or with seine nets from the shore and euthanized with an overdose of benzocaine prior to fixation in ethanol. Species identification was based on morphological characters (patterns of sensory papillae and pores, coloration, squamation) using the identification keys of Miller (1986) and Kovačić (2008) as well as the original descriptions for more recently described species (i.e., Kovačić & Šanda, 2016 for G. incognitus;Miller & Šanda, 2008 for P. montenegrensis). The sampling complied with protocols approved by the responsible governmental authorities (according to the countries where the sampling was conducted). The number of specimens obtained varied from species to species, owing to differences in population sizes or difficulty of access. It is generally recommended that at least 30 specimens per category should be used in a statistical Downloaded from Brill.com11/03/2020 07:41:22AM via free access analysis (Kar & Ramalingam, 2013). However, this criterion is an arbitrary one and cannot always be fulfilled in organismic studies, as in the present instance. In such a case, a minimum of seven specimens per species is often considered to be sufficient for a statistical analysis in fish biology. Here, we compiled for each genus a complete dataset containing all collected species, and a reduced dataset including only those species for which at least seven specimens were available. This resulted in four datasets (two for each genus), with 14 and ten species, respectively, in the case of Gobius, and seven and five species, respectively, in the case of Pomatoschistus. Each dataset was subjected to univariate statistical analyses and assessment of Fourier descriptors. For Linear Discriminant Analysis (LDA), only the reduced datasets were used (see below).

Otolith preparation and SEM imaging
Fish skulls were opened dorsally under a microscope, and the left and right saccular otoliths ("otolith" hereafter) were removed. Otoliths were incubated in 5% KOH solution for 1 h to clean off adherent organic matter, immersed in distilled water for 1-2 h, then washed several times with distilled water, and stored in the dry state. All otoliths were mounted and gold-coated for scanning electron microscopy (with either a SEM LEO 1430 VP at ZSM Munich, or a HITACHI SU 5000 Schottky FE-SEM at LMU Munich). All otoliths are kept in the Bavarian State collection Palaeontology and Geology under the inventory number 2020 LIV.

Otolith morphometry
Ten measurements were taken from the SEM images of the left otolith of each specimen following Gierl et al. (2018) and using ImageJ2 (Rueden et al., 2017). The parameters chosen were maximum length, maximum height, perimeter length, areas of both the whole otolith and the sulcus, as well as the vertical distances from the anterior and posterior ends of the sulcus to the ventral margin of the otolith ( fig. 1c) Adriatic Sea Table 1 Details of the sample sets of the Gobius and Pomatoschistus species studied (Cont.) Abbreviations: SL, standard length; N, number of fish specimens and otoliths; *ten specimens were used for Fourier analysis, for the other analyses seven specimens were used (see supplementary tables S1-S5).
Downloaded from Brill.com11/03/2020 07:41:22AM via free access et al., 2018), which were multiplied by 1000 for subsequent statistical analyses. The original measurements and the otolith variables derived from them are tabulated for all specimens in supplementary tables S1 and S2.

Otolith shape analysis
On the basis of the SEM images of both right and left otoliths, each otolith image was transformed into a black object on white background using Photoshop CS6 and saved as a JPG file. To conduct elliptic Fourier shape analysis, JPG-files were imported into R using the Momocs software package (Bonhomme et al., 2014). After Momocs had extracted the outlines of the otoliths, the appropriate sequence of commands was implemented to normalize the perimeter outlines in terms of size and orientation, and to estimate the number of harmonics required to attain a cumulative power that corresponded to 99.9% of the shape (24 harmonics in Gobius and 20 harmonics in Pomatoschistus). Subsequently, an elliptic Fourier analysis was performed, which resulted in the Fourier descriptors (supplementary table S3). As no significant intraspecific differences in shape were detected between left and right otoliths (MANOVA, P < 0.05), SEM images of both were included in the subsequent analyses, as has been done in previous work (Davoodi & Rahimian, 2016).

Meristic counts
Meristic characters were counted based on the X-ray images. They include number of abdominal, caudal, and total vertebrae (last terminal vertebral element was included); number of fin elements in second dorsal fin (D2) and anal fin (A) (the two rays attached to the last pterygiophore were counted as one); number of anal fin pterygiophores inserting in front of haemal spine of first caudal vertebra (AP, Birdsong et al., 1988); and number of dorsal and ventral procurrent rays ( fig. 1b). The number of fin elements (spinous rays) in the first dorsal fin (D1) and the number of branched caudal rays did not contain useful information as no differences were found in either the Gobius or the Pomatoschistus dataset. Each meristic count was used as a single variable in the statistical analyses. Counts of all specimens are provided in the supplementary table S5.

Statistical analysis
Variables resulting from otolith morphometry and Fourier shape analyses, body morphometry, and meristic characters each served as input for statistical analyses in R (R Core Team, 2019), using the packages Momocs, car, stats, PMCMRplus, MVN and heplot, as well as PAST Figure 1 a-b, Schematic drawing of a goby skeleton depicting the standard morphometric (1a) and meristic (1b) characters used in this study; c-d, Schematic drawing (c) and SEM image (d) of a right goby otolith (based on G. bucchichi) showing the measured distances and areas (after Gierl et al., 2018)  vers. 3.22 (Hammer et al., 2001). Normal distribution was tested based on the Shapiro-Wilk test (P > 0.05 if normal distribution is satisfied). Homogeneity of variances was tested using Levene's test (P > 0.05 if homogeneity is satisfied) for the morphometric variables, and the Fligner-Killeen test (P > 0.05) for the meristic variables. Normality and homogeneity of the Fourier descriptors were examined with Mardia's test (skewness and kurtosis, P > 0.05). One-way analysis of variance (ANO-VA, P < 0.05) was used to evaluate whether any morphometric variable was significant for a species. Some otolith morphometric variables revealed covariance with OL (Spearman, P < 0.05); in this case one-way analysis of covariance (ANCOVA, P <0.05), with otolith length as covariable was applied. If homogeneity was fulfilled, Tukey's HSD post-hoc test (P < 0.05) was conducted; if homogeneity was not given we carried out Dunnett's T3 post-hoc test (P < 0.05). A Kruskal-Wallis test (P < 0.05) with Dunn's post-hoc test (P < 0.05) was performed to ascertain whether a given meristic value was significant for a species. Furthermore, the otolith and body morphometric datasets, and the meristic variables were each subjected to Principal Component Analysis (based on the individual values per specimen and species). The resulting first four principal components (PC1-4) were inspected for their significance at species level using ANOVA with Tukey's HSD post-hoc test (P < 0.05). The first two PCs of the Fourier descriptors were used for a multivariate analysis of variance (MANOVA) (P < 0.05). Based on the reduced datasets for Gobius and Pomatoschistus, respectively, a LDA was executed based on PC1-4 of the different arrays of variables derived from the otolith and body morphometry (Box's M Test, P > 0.05). The maximal separation power of the Fourier descriptors without overfitting the model yielded 37 PCs in case of Gobius and 14 PCs in case of Pomatoschistus. Accordingly, the LDA was conducted based on PC1-37 (Gobius), and PC1-14 (Pomatoschistus) of the Fourier descriptors (Box's M Test, P > 0.05). The classification success of each LDA is given based on jack-knifed cross-validation.

Evaluation of characters to discriminate between extant goby species
For our quantitative approach, we evaluated the taxonomic value of each variable by calculating its separation success for each species. In other words, we measured from how many of its congeners a given species could be separated based on a certain variable. Thus, 100% separation success means that the variable concerned could separate a species from all of its congeners (included in the respective dataset) and 0% indicates that it could not separate that species from any of its congeners. Moreover, to minimize the impact of potentially coincidental results, we applied the following constraint: A given variable was considered to be taxonomically significant if it discriminated between at least two species, and could also differentiate each of those two species from at least 50% of the congeners (in the respective dataset). Variables that fulfilled this constraint are depicted against a green background in the relevant tables (see below). Variables that failed this test, but were indicative for one species with a success rate ≥50% are indicated in italics.

Otolith descriptions
High-quality SEM images of otoliths obtained from extant goby species are rarely found in previous publications. Therefore, in addition to the morphometric approaches, SEM images of representative otoliths of each species included in our study are presented and brief descriptions for the otoliths of each species are provided. Otolith terminology follows Nolf (1985) and Lombarte et al. (2018) (see here fig. 1d).

Abbreviations
For otolith morphometry: OA, otolith area; OH, otolith height; OL, otolith length; OP, otolith perimeter; SuA, sulcus area; SuH, sulcus height; SuL, sulcus length; SuP, sulcus perimeter; SuEndV, vertical distance from posterior end of sulcus to ventral margin of otolith; Su-TipV, vertical distance from anterior end of sulcus to ventral margin of otolith For body morphometry: Ab, length of anal fin base; B, body depth at origin of first dorsal fin; CP, length of caudal peduncle; D2b, length of second dorsal fin base; D2C, distance between end of second dorsal fin and first dorsal (procurrent) ray of caudal fin; SL, standard length (from snout to begin of branched caudal fin rays); SN/A, distance from snout to origin of anal fin; SN/D1, distance from snout to origin of first dorsal fin; SN/D2, distance from snout to origin of second dorsal fin; TL, total length. All morphometric measurements are indicated in fig. 1a.

Species separation in Gobius
Otolith morphometry Using the complete dataset (N, 14 species), ten otolith variables (out of 23) were found to be useful for species separation. These variables were significant for a total of eight species (ANCOVA/ANOVA, table 2a1). The highest separation success was achieved with the variables SuL/SuH (77% for G. geniporus) and SuP/SuEndV (77% for G. paganellus) ( table 2a1). The five most successful variables were OL/OH, SuL/OP, SuL/SuEndV, SuH/ SuEndV and SuP/SuEndV, but each of these separated no more than three of the 14 included species (table 2a1). ANOVA based on PC1-4 calculated from the otolith variables successfully discriminated between five species (table 2b1). Gobius vittatus was separated from 69%, the others from 54% of the congeners. Notably, G. cruentatus, G. gasteveni and G. roulei, which had not been separated when the individual otolith variables were used, could now be distinguished from each other (compare table 2b1 vs. 2a1).
When the reduced dataset (N, 10 species) was used, 17 otolith variables (out of 23) were successful in species separation (ANCOVA/ ANOVA, table 2a2). The same two variables as in the complete dataset (SuL/SuH, SuP/ SuEndV) had maximum separation success and also discriminated the same species (G. geniporus, G. paganellus), each with increased success (89% vs. 77%). Six further variables were also taxonomically indicative, as each discriminated one or two species from all but two of the congeners (78% success). These variables were OL/OH (G. incognitus), OP/OL (G. cobitis), SuP/OP (G. vittatus), SuL/ OP (G. auratus, G. vittatus), SuL/OL (G. auratus), SuL/OH (G. incognitus), and SuL/SuEndV (G. auratus) (table 2a2). With respect to the power of a given variable to discriminate between many species, the two most efficient variables were SuL/SuEndV and SuH/SuEndV; each of these could discriminate between six species (vs. three species when the complete dataset was used). In total, all species could be discriminated in the reduced dataset from ≥50% of the congeners (table 2a2). Four species were separated by eight to ten Downloaded from Brill.com11/03/2020 07:41:22AM via free access Table 2 Results of the statistical analyses based on the Gobius datasets and the indicated variables. The success of a certain variable in separating a species is indicated in % (100% means that the species is separated from all congeners by this variable). Light / dark green color denotes that at least two species were separated by the given variable, and each of those two species from ≥ 50% / ≥ 80% of the congeners (in the respective dataset)  Only five species were separated in this case and separation success for a given species varied from 54 to 62% (table 2e1). Gobius cruentatus and G. geniporus, which could not be distinguished in the preceding analysis, were now separated. The other three species that could be discriminated were G. bucchichi, G. incognitus and G. vittatus.
The results based on the reduced dataset (table 2d2, e2) were very similar to those obtained from the complete dataset. The same four morphometric variables could separate a species (SN/D1, SN/D2, D2b, Ab), and the individual species were separated with the same success (56-78%; see table 2d2). As before, most powerful in relation to the number of species that could be separated were Ab (four species) and D2b (five species). With the single exception of G. roulei (not separated), the same species as in the complete data were separated (five in total), each based on one to four variables (table  2d2). When PC1-4 of the morphometric variables were used in the ANOVA, a total of seven species could be separated (  . 2a). The members of each pair show some overlap, but only relatively moderate overlap with the other species considered.
The first two functions of the LDA based on PC1-23 of the otolith Fourier descriptors captured 39.1% and 19.1% of the variation, respectively ( fig. 2b). Overall classification success (jack-knifed) of this LDA was 85% (table 3b). 100% separation success was achieved for G. cruentatus and G. geniporus (vs. 18% and 60% in the previous LDA), and 95% was obtained for G. cobitis (vs. 50% in the previous LDA) ( Table 3b). The scatter plot delineates three groups: Gobius cobitis, which had broadly overlapped with G. cruentatus in the preceding LDA, is now distinct (fig. 2b). The second group includes G. geniporus, G. niger and G. roulei, and the third one comprises the remaining species (fig. 2b).
The first two functions of the LDA based on the body morphometric variables accounted for 66.3% and 18.6% of the variation, respectively. Overall classification success (jackknifed) was 54.6% (table 3c). Classification success was highest for Gobius vittatus and G. bucchichi, with 81.8% and 70%, respectively, while the success rates were lowest for G. auratus, G. cobitis and G. roulei (27-30%) (table 3c). The scatter plot revealed overlap of almost all species, except G. vittatus, which is comparatively distinct, and G. bucchichi and G. incognitus, which overlap with each other, but not or not very much with the remainder (fig. 2c).

Species separation in Pomatoschistus
Otolith morphometry Five otolith variables (out of 23) were successful in species separation when the complete dataset was used (N, 7 species; ANCOVA/   Overall classification success (jack-knifed): 85.0% Table 3 Results of the Linear Discriminant Analysis (LDA) using the reduced dataset of Gobius ANOVA; table 4a1). The maximum classification success for a given species was 67%. OL/ OH and OP/OH were the best variables with respect to the number of species that could be discriminated; OL/OH separated all species except P. knerii, and OP/OH discriminated four species (table 4a1). In total, all species could be separated from one another; P. marmoratus, P. microps and P. montenegrensis each by four variables, and the remainder by a single variable each (table 4a1). When PC1-4 calculated from the otolith morphometric variables was analysed, the success in species separation declined in the case of P. microps (33% vs. 67% in the preceding analysis) and P. pictus (0 vs. 50%), but was improved in the case of P. montenegrensis (83% vs. 67%; table 4b1). When the reduced dataset (N, 5 species) was analysed in the same way, most of the otolith morphometric variables contributed to species separation (table 4a2). Four variables achieved 100% separation success for a given species: SuEndV/OP for P. marmoratus and P. microps, and SuH/OL, SuH/SuTipV and SuP/SuTipV each for P. montenegrensis (table  4a2). Overall, all species could be separated from ≥50% of their congeners, and 13-17 variables contributed in each case (table 4a2). As observed in the complete dataset, the results of ANOVA on the basis of PC1-4 of the otolith morphometric variables discriminated the species less efficiently than when the individual otolith variables were used (compare table 4b2 vs. 4a2).
Fourier shape analysis of the otoliths When the complete dataset was used, PC1 and PC2 of the Fourier descriptors attained success rates in species separation ranging from 67% (P. microps, P. pictus) to 100% (P. knerii, P. montenegrensis, P. quagga) (MANOVA;     Body morphometry Each of the eight morphometric variables contributed to species separation and each species was separated from all others when the complete dataset was used (ANOVA; table 4d1). 100% separation success was achieved by CP (for P. minutus) and Ab (for P. montenegrensis). The most efficient variables in relation to the number of species discriminated were Ab (all species), and D2b, D2C and SN/A (six species). Pomatoschistus quagga was the only species that could be separated by each morphometric variable from 50-83% of its congeners; the remaining species were separated by three to six variables from their congeners (table 4d1). Furthermore, each species was separated when PC1-4 of the morphometric variables were analysed (ANOVA; table 4e1). 100% separation success was obtained for P. montenegrensis and P. quagga (table  4e1).
Overall, similar results were obtained with the reduced dataset, but some differences are noteworthy (table 4d2). With the exception of SN/D2, every morphometric variable could be used to separate some species. The reason that SN/D2 was now disqualified lies in the exclusion of P. minutus (which had been separated by this variable) from the reduced dataset. Three variables (vs. one in the complete dataset) showed 100% separation success for at least one species: D2C (for P. marmoratus, P. microps), B (for P. quagga) and Ab (for P. montenegrensis). Most effective concerning the number of species that could be separated were D2C (all species) and SN/A (four species) (table 4d2). Apart from P. marmoratus (separated by a single variable), each of the included species could be discriminated by five or six variables from their congeners (table 4d2). When ANOVA was done based on PC1-4 of the morphometric body variables, the separation success was generally similar to that of the previous analysis and all species could be separated (table 4e2).

Meristic characters
Numbers of total and abdominal vertebrae contributed to species separation with moderate success (50-67%) when the complete dataset was employed (table 4f1). Pomatoschistus microps, P. montenegrensis and P. pictus could be separated by either the total or the abdominal vertebrae count, while P. quagga was discriminated by both of these counts (table 4f1). ANOVA using PC1-4 of the meristic counts showed improved results, as PC1 separated all species apart from P. pictus, with highest separation success (83%) for P. minutus, P. montenegrensis, and P. quagga (table 4g1).
In the reduced dataset, seven meristic characters contributed to species separation (vs. two in the complete dataset), with an improved overall separation success (50-100% vs. 50-67%) (table 4f2). The caudal and the abdominal vertebrae count separated P. montenegrensis and P. microps/P. quagga, respectively, from all other species. The latter was most efficient in relation to the number of species that could be separated (all five species) (table 4f2). In total, all five species could be separated: P. knerii, P. montenegrensis and P. quagga by six, seven and five variables, respectively, and P. marmoratus and P. microps by one and three variables, respectively (table 4f2). ANOVA using PC1-4 of the meristic values separated each species from 75-100% of its congeners, with 100% success for P. microps, P. montenegrensis and P. quagga (table 4g2).

Discriminant analyses
As for Gobius, each LDA was based on the reduced dataset. The variables used as input were (i) PC1-4 of the otolith morphometric variables, (ii) PC1-14 of the Fourier descriptors, and (iii) PC1-4 of the body morphometric variables (table 5).
The first two functions of the LDA based on PC1-4 of the otolith morphometric variables explained 62.0% and 33.2% of the variation, respectively ( fig. 2d). Overall classification success (jack-knifed) of the LDA was 71.1%. The highest classification success was achieved for P. montenegrensis (90%) and P. microps (87%), while the success rates were between 56 and 60% for P. knerii, P. marmoratus, and P. quagga (table 5a). The scatter plot depicts three groups ( fig. 2d). The first consists of P. montenegrensis and is completely distinct. The second and third groups include P. microps/P. marmoratus and P. knerii/P. quagga; the members of each group overlap with each other, but very little or not at all with the others (fig. 2d).
The first two functions of the LDA based on PC1-14 of the Fourier descriptors captured 57.6% and 21.6% of the variance, respectively. Overall classification success (jack-knifed) was 96%, with 100% success for P. microps and P. quagga, >90% success for the remainder (table 5b). The corresponding scatter plot depicts P. knerii, P. marmoratus, and P. montenegrensis as separate groups, while P. microps and P. quagga overlap ( fig. 2e).
The first two functions of the LDA based on the body morphometric variables account for 71.7% and 23% of the variation, respectively. Overall classification success (jackknifed) was 75.9%, with 90% success for P. montenegrensis and P. quagga, and 67-70% success for the remainder (table 5c). The success in separation is also seen in the scatter plot, which reveals that each species is relatively well separated, although some overlap is seen ( fig. 2f).

Summary of the results based on the statistical analyses
A comparison of the results based on the four methodological approaches (otolith morphometry, otolith Fourier shape analysis, body morphometry, meristic counts) reveals that Fourier shape analysis of otoliths is the most powerful species discriminator of all the methods tested here (table 6). A plot of the relative discriminatory power of the individual variables used in the other three approaches indicates that rates of success in species separation vary, and implies that some characters have more taxonomic significance than others (table 7). For example, none of the individual otolith variables had the power to separate all of the species of Gobius or Pomatoschistus with good success (i.e., with ≥75%; table 7). While some variables (SuTipV/ OP, SuH/SuEndV) were capable (albeit with moderate success) of separating some of the species of Gobius from the others, these same variables failed to discriminate between species of Pomatoschistus spp., and vice versa (table 7). Furthermore, a single variable could be indicative for a certain species, but be unable to reliably discriminate its congeners. An example is SuL/SuH, which achieved good separation success only for G. geniporus (89%) and G. paganellus (78%) (reduced dataset, table 2a2).
A further outcome was that the number of species that could be separated from their congeners with a success of ≥50% declined when PC1-4 of the otolith or the morphometric variables were analysed instead of the individual variables (table 6, complete datasets).
Moreover, the number of specimens employed in the statistical analyses did not necessarily influence their separation success. Examples are G. fallax (represented with only two specimens) and G. gasteveni (one specimen), which were well separated from their congeners using body morphometry, whereas G. paganellus and G. roulei (ten specimens each) could not be separated based on the same approach (table 2d1). On the other hand, as already pointed out by Lombarte et al. (2018), the number of species used in the statistical analysis could have an influence on the results, insofar as species separation becomes more successful when the number of species considered decreases, and less successful when the number of species increases. This Table 5 Results of the Linear Discriminant Analysis (LDA) using the reduced dataset of Pomatoschistus and the indicated variables. The numbers in rows are percentages that denote the classification into the species given in columns (correctly classified species are highlighted in grey)  Otolith variables (AN(C)OVA) 8 (57%) All All All PC1-4 of otolith variables (ANOVA) 5 (36%) 7 (70%) 5 (71%) 4 (80%) PC1-4 of otolith variables (LDA) -6 (60%) -All Fourier otolith shape analysis (MANOVA) All All All All Fourier otolith shape analysis (LDA) -All -All Fish morph. variables (ANOVA) 9 (64%) 5 (50%) All All PC1-4 of fish morph. variables (ANOVA) 5 (36%) 7 (70%) All All PC1-4 of fish morph. variables (LDA) -6 (60%) -All Meristic variables (Kruskal-Wallis) 0 2 (20%) 4 (57%) All PC1-4 of meristic variables (ANOVA) 3 (21%) 3 (30%) 6 (86%) All phenomenon appears to be related to growing spread (variance) after inserting values of additional species. In other words, the additional values increased the overall variance of the variables, which finally blurs the information of a certain variable that would otherwise be characteristic for an individual species. This likely explains why in our study the success of a given variable in species separation generally increased when the reduced datasets of Pomatoschistus and Gobius were used (tables 6, 7).

c Classification success (jack-knifed) based on LDA using PC1-4 of fish morph. variables
△, at least two species were separated from ≥50% of their congeners; ▲, ≥50% of the species included were separated from ≥50% of the congeners. See methods for abbreviations of variables.   . 3), (ii) the otoliths of Gobius species that could be identified based on otolith variables ( fig. 4), (iii) the otoliths of Gobius species that could not be identified based on otolith variables ( fig. 5), and (iv) the otoliths of the species of Pomatoschistus ( fig. 6). The descriptions refer to the mesial side of the otoliths, which, as in all saccular otoliths of teleost fish, bears the sulcus and most relevant characters. In all studied otoliths, the sulcus acusticus is slightly inclined and reveals the 'shoe solelike' shape that is typical for most Gobioidei (Nolf, 1985(Nolf, , 2013Gierl et al., 2018;Lombarte et al., 2018). The cauda is slightly or clearly shorter than the ostium and the ostium is anteriorly tapering or pointed (figs. 3-6). The sulcus of all species is covered by a microcrystalline texture that is usually smooth, with exceptions in G. niger and G. roulei (presence of some flat bumps; fig. 5g-l), and G. cobitis, G. incognitus and G. paganellus (lack of this texture along the dorsal part of the ostium; fig. 4a-c, h-k). Some irregularities of the sulcus shape can occur within a species: one specimen of G. couchi showed a triangular extension of the dorsal ostial margin ( fig. 3f), and one specimen of G. fallax revealed an irregular shape of the ostium (fig. 3h). In the otoliths of Pomatoschistus, the shape of the sulcus is generally somewhat variable ( fig. 6). Additionally, the size of the cauda may be reduced in P. knerii ( fig. 6a-d), P. marmoratus ( fig. 6e-h) and P. microps ( fig. 6i-l), and is always reduced in P. montenegrensis ( fig. 6n-p) and P. quagga ( fig. 6r-t). A well defined, rounded or elongate, thick crista inferior is visible at the beginning of the cauda in almost all species of Gobius and Pomatoschistus ('subcaudal iugum' sensu Schwarzhans, 2014). The shape and strength of the crista inferior can vary between and also within a species (figs. 3-6). In otoliths of Pomatoschistus that possess a reduced cauda, the crista inferior reaches almost the end of the cauda (P. knerii, P. marmoratus, P. microps; fig. 6a-l), or even continues beyond (P. montenegrensis, P. quagga; fig. 6n-p, r-t).
The Gobius auratus species complex ( fig. 3) The otolith shape in all species of the G. auratus species complex is approximately rhomboidal, with the highest shape seen in G. couchi ( fig. 3d-f), and the longest in G. kolombatovici ( fig. 3k, l). All species display a prominent posterodorsal projection at the posterior margin and slight differences in the shape of this projection occur between the species: it is blunt in G. auratus ( fig. 3a-c) and G. fallax ( fig. 3g, h), slightly tapering in G. couchi ( fig. 3d-l), short and truncated in G. gasteveni ( fig. 3i, j), and broadly rounded in G. kolombatovici ( fig. 3k, l). The further posterior margin is separated from the posterodorsal projection by a clear incision or notch; it is slightly rounded in G. auratus and G. fallax, and relatively straight in the remainders. The ventral margin is slightly curved in G. couchi and G. gasteveni, in the others it is relatively straight. A well developed preventral projection occurs in G. fallax, G. gasteveni, G. kolombatovici and, to a lesser extent, also in G. couchi. In G. auratus no clear preventral projection is present, but the ventral margin ends in a sharp point. The anterior margin is usually concave or slightly incised in the middle in all species. A small predorsal projection can be developed, but its size and shape is variable within and between the species. It seems that otoliths of smaller specimens tend to have such a projection, while those of larger specimens do not (see fig. 3). The dorsal margin is rounded and can bear some lobes or crenulations; only Selce, 5l; c: Krk, 2r mirrored), G. couchi (d-f: Krk, 2l, 3r mirrored, 1l), G. fallax (g, h: Unije Island, 2r mirrored, 1l), G. gasteveni (i, j: Galicia, 1l, 1r mirrored) and G. kolombatovici (k, l: Krk, 1l, 2l). Numbers following the localities refer to the fish specimen from which the otolith was extracted; l, left otolith; r, right otolith, mirrored for better comparison. SL denotes the standard length (in mm) of the corresponding fish specimen. Scale bars: 0.5 mm. All figured otoliths are kept in the Bavarian State Collection (collection number SNSB-BSPG 2020 LIV).  niger (g-i: Pilsey Island, 2l, 6l, 5l) and G. roulei (j-l: Selce, 2l, 1l, 3l). Numbers following the localities refer to the fish specimen from which the otolith was extracted; l, left otolith. SL denotes the standard length (in mm) of the corresponding fish specimen. Scale bars: 0.5 mm. All figured otoliths are kept in the Bavarian State Collection (collection number SNSB-BSPG 2020 LIV). Figure 6 Otoliths (mesial view) of Pomatoschistus knerii (a-d: Krk, 1r, 2r, each mirrored, 9l, 4r mirrored), P. marmoratus (e-h: Selce, 1r, 2r, 3r, each mirrored, 4l), P. microps (i-l: Stralsund, 8l, 5l, 6l, 14l), P. minutus (m: Stralsund, 1l), P. montenegrensis (n-p: Skadar lake, 8l, 9l, 6l), P. pictus (m: Norway, 2r mirrored), and P. quagga (r-t: Krk, 2l, 8r mirrored, 4l). Numbers following the localities refer to the fish specimen from which the otolith was extracted; l, left otolith; r, right otolith, mirrored for better comparison. SL denotes the standard length (in mm) of the corresponding fish specimen. Scale bars: 0.5 mm. All figured otoliths are kept in the Bavarian State Collection (collection number SNSB-BSPG 2020 LIV).
in G. kolombatovici it is relatively flat. A median tip of the dorsal margin is visible in G. auratus and G. couchi.
Gobius species that could be identified based on otolith variables ( fig. 4) Six species could be recognized with good success based on their otolith variables (table 2a), i.e., G. auratus (described above), G. cobitis, G. geniporus, G. incognitus, G. paganellus and G. vittatus. Of those, the otoliths of G. incognitus ( fig. 4g-i) and G. vittatus ( fig. 4m-p) display a rhomboidal shape, while the otoliths of G. cobitis ( fig. 4a-c), G. geniporus ( fig. 4d-f), and G. paganellus ( fig. 4j-l) are more rectangular. The otoliths of all five species are usually longer than high except some otoliths of G. vittatus, which are approximately quadratic. All five species possess a prominent posterodorsal projection on the posterior margin, which differs slightly between them: it is blunt, slightly to distinctly crenulated and mostly facing upwards (dorsally) in G. cobitis ( fig. 4a-c) and G. paganellus ( fig. 4j-l); it is slightly pointed and mostly facing upwards in G. incognitus ( fig. 4g-i); it is of triangular-rounded shape in G. vittatus ( fig. 4m-p); and it is broadly rounded in G. geniporus ( fig. 4d-f). The posterior margin beneath the projection is relatively straight in all five species and can bear slight crenulations, only in G. vittatus it is always smooth. The ventral margin is faintly curved in G. incognitus and G. vittatus, and relatively straight in G. cobitis, G. geniporus, and G. paganellus. A clear preventral projection occurs only in some otoliths of G. cobitis (fig. 4c) and G. paganellus ( fig. 4k, l). In both G. incognitus and G. vittatus the anterior end of the ventral margin is relatively sharp and pointed. The anterior margin is straight in G. cobitis, concave in G. geniporus, oblique in G. incognitus and G. paganellus, and straight with a clear incision in the middle in G. vittatus. A predorsal projection occurs only in G. geniporus and G. vittatus, it is sharp in G. geniporus, while it is rounded in G. vittatus. A relatively flat dorsal margin occurs in G. incognitus and G. paganellus; the dorsal margin is regularly rounded in the other three species. A median dorsal tip is usually visible in G. vittatus and in some otoliths of G. incognitus. A smooth dorsal margin occurs in G. vittatus, in the other species the dorsal margin can bear lobes or crenulations.
Gobius species that could not be identified based on otolith variables (Fig. 5) The otoliths of G. bucchichi, G. cruentatus, G. niger, and G. roulei could not be separated based on the individual otolith variables (table 2a). Their otoliths are longer than high, with a clear rhomboidal shape in G. bucchichi ( fig. 5a-c), and a rectangular to rhomboidal shape in G. cruentatus ( fig. 5d-f), G. niger ( fig. 5g-i), and G. roulei ( fig. 5j-l). The otoliths of all four species possess a clear posterodorsal projection, which reveals slight differences between the species: it is rounded to slightly pointed in G. bucchichi; it is rounded-tosquared in G. cruentatus; and it is broad, mostly rounded and sometimes slightly lobed in G. niger and G. roulei. The posterior margin beneath the projection and also the posteroventral angle are clearly (G. bucchichi) or slightly (G. roulei) rounded. In G. cruentatus the posterior margin is straight and can be weakly crenulated, in G. niger it is straight or curved; the posteroventral angle is mostly angular in these two species. The ventral margin is straight in G. cruentatus, and faintly curved in the other three species. A weak preventral projection occurs in some otoliths of G. cruentatus (fig. 5d) and G. roulei ( fig. 5j-l). The anterior margin is mostly concave, with an incision in the middle part. An angular predorsal angle is typical for G. bucchichi, G. cruentatus and G. niger; in G. roulei a rounded predorsal "angle" can occur ( fig. 5k, l). The dorsal margin is mostly slightly undulated; it is gently inclined in G. bucchichi, slightly rounded in G. cruentatus, clearly rounded in G. roulei, and strongly rounded in G. niger.
The otoliths of the Pomatoschistus species ( fig. 6) The otoliths of Pomatoschistus are usually rounded to square, but reveal a relatively high intraspecific variation regarding the details of their contours (fig. 6). The otoliths of P. knerii ( fig. 6a-d), P. montenegrensis ( fig. 6n-p) and P. quagga ( fig. 6r-t) are slightly higher than long, the otoliths of P. minutus reveal a relatively rectangular shape ( fig. 6m), and those of P. marmoratus ( fig. 6e-h) and P. microps ( fig. 6i-l) are nearly quadratic. A posterodorsal projection can be present or not in P. marmoratus, P. microps and P. minutus, it is lacking in P. knerii, P. microps and P. pictus, and it is usually not present or replaced by a broad bulge in P. montenegrensis and P. quagga. The further posterior margin is relatively straight and slightly or clearly incised approximately in the middle; the posteroventral angle is rounded, angular or oblique. The ventral margin is straight or faintly curved in all species, no preventral projection is present. The anterior margin is straight, oblique or rounded, an incision in the middle may be present or not. An angular predorsal angle may occur in all species, except in P. knerii. The rounded dorsal margin is slightly lobed or crenulated in P. marmoratus, P. microps and P. minutus. It is smooth in P. knerii, P. pictus, P. montenegrensis, and P. quagga; a clear median tip can occur in the latter two ( fig. 6n, p, r).

Discussion
The objective of this study was to evaluate the ability of characters of otoliths, body morphometric traits and meristic counts, all of which can be preserved in fossils, to discriminate species from their congeners.

Evaluation of otolith morphometry and otolith shape analysis
The morphology of the sagittal otolith is generally considered species-specific. This view goes back to the pioneering work of Koken (1884Koken ( , 1891 and has since been demonstrated in numerous studies dealing with many different fish groups (Nolf, 1985;Campana, 2004;Tuset et al., 2008). Besides genetic and ontogenetic factors, also environmental parameters can influence the overall otolith shape (Tuset et al., 2003;Volpedo & Echeverría, 2003;Vignon & Morat, 2010). Regarding otolith morphometry, the varying results from different statistical analysis (table 6), and the fact that the taxonomic significance of a certain otolith variable depends on the species under consideration are consistent with previous work on the taxonomic value of otolith morphometry (e.g., Reichenbacher & Reichard, 2014;Avigliano et al., 2016;Gierl et al., 2018). In the context of samples of fossil goby otoliths and attempts to detect the number of species represented, it therefore may be worth considering all otolith variables at the outset of the study.
Otolith shape analysis is an established method for stock or population discrimination in fisheries research (Campana & Casselmann, 1993;Mérigot et al., 2007;Stransky et al., 2008;Libungan et al., 2015), but not commonly used for species discrimination. But for gobies, several studies are available that have analysed the differences between species: Elliptic Fourier analysis was conducted by Lord et al. (2012) to analyse three sympatric species of Sicyopterus, by Bani et al. (2013) to examine three sympatric Caspian species (one species of Neogobius, two of Ponticola), by Davoodi & Rahimian (2016)  and by Yu et al. (2014) for the study of five species representing five different genera from northern Chinese coastal waters. Moreover, Lombarte et al. (2018) used wavelet analysis to examine 25 Mediterranean goby species distributed in 14 genera (including ten species of Gobius and two species of Pomatoschistus). In each of these studies, the authors found that the species included could be reliably discriminated based on their otoliths, but that the classification success varied among species. Our study largely confirms these results. In addition, it significantly advances our knowledge of the interspecific differences between goby otoliths within a given genus, as we focused on the discrimination of individual species from their congeners. Moreover, like Bani et al. (2013) and Yu et al. (2014) we found that otolith Fourier shape analysis works more effectively in goby species separation than otolith morphometry. This appears to be related to the circumstance that otolith shape differences between goby species can be very subtle (see figs. 3-6), and in such cases cannot easily (or not at all) be quantified using single otolith variables (see Yu et al., 2014;Bonhomme et al., 2014;Caillon et al., 2018). The high discriminatory power of the otolith Fourier shape analysis makes this method a promising tool to examine ancient species diversity when abundant and well preserved fossil otoliths are present.

Evaluation of body morphometry
Body morphometry in fish can be greatly influenced by the demands of swimming, and associated factors like habitat size or presence/absence of predators and competitors (Gholami et al., 2015 and references therein). Therefore, body morphometry is often not considered to be relevant for species identification. Also, the taxonomic keys to the species of Gobius and Pomatoschistus provided in Miller (1986) do not include morphometric characters. On the other hand, body morphometry can be very efficient in detecting differences between populations or stocks (e.g., Haddon & Willis, 1995;Turan, 2004;Cheng et al., 2005;Buj et al., 2008). In our study, some body morphometric variables showed a high degree of separation success among species of both Gobius and Pomatoschistus (tables 2d, 4d, 7); but only one population per species was available. Addition of further populations could reveal intraspecific differences in some of these variables, and eventually render them unsuitable for species separation. However, fossil finds are usually restricted to a single location, and often to a certain timeframe. Perhaps the most famous example is the locality of the fossil Lagerstätte Monte Bolca (c. 50 Ma; Marramà et al., 2016 and references therein). Examples of fossil locations bearing several goby fossils include Illerkirchberg in Germany (c. 17 Ma; Gierl & Reichenbacher, 2015), Klinci in Serbia (c. 16-19 Ma;Bradić-Milinović et al., 2019) and Oran in Algeria (c. 5-6 Ma; Arambourg, 1927). Our results indicate that measurements of SN/D1, SN/D2, D2b, and Ab are well suited to evaluating the ancient goby species diversity found in such singular localities.

Evaluation of the meristic counts
Meristic traits are usually estimated to work better than morphometry in species separation, but meristic counts among gobies are often very similar (Birdsong et al., 1988). In our study, the meristic values had little or no discriminatory power within Gobius, whereas they worked more efficiently for Pomatoschistus, especially when the reduced dataset was used (tables 6, 7). It is possible that the use of only five species in the reduced Pomatoschistus dataset contributed to the greater separation success in that case (see above and Lombarte et al., 2018). However, the most conspicuous meristic character that distinguishes Pomatoschistus from Gobius is the elongated vertebral column, usually comprising 30-31 vertebrae (vs. 28 in Gobius) (Miller, 1986;supplementary table S3). An elongated body form can, among other things, improve swimming abilities, affect feeding habits, and promote adaptation to specialized microhabitats (Claverie & Wainwright, 2014). An example among the Oxudercidae is Luciogobius, whose adaptive radiation into interstitial microhabitats was facilitated by marked body elongation (Yamada et al., 2009). Thus, it is possible that species of Pomatoschistus can be more readily separated from each other based on meristic counts than species of Gobius because they are adapted to more specific microhabitats. However, as data on microhabitats such as distinctive grain sizes, or abundance of certain algae, or presence of distinct shell sizes for spawning, are relatively limited for species of Pomatoschistus and Gobius (Miller, 1986;Nellbring, 1993;Wilkins & Myers, 1992;Kovačić & Šanda, 2016), this issue cannot be resolved at present. In the context of fossils, our results indicate that shared meristic characters among fossil gobies need not necessarily indicate the presence of only a single species, as has been proposed in some previous work (e.g., Gaudant & Quayle, 1988;Gaudant, 1998;Reichenbacher et al., 2007).

Comparisons of the otolith morphology of
Gobius with previous works Generally, digital photographs or SEM images of extant species of Gobius are rarely provided in the literature and the otoliths of G. fallax ( fig. 3g, h), G. gasteveni ( fig. 3i, j), G. kolombatovici ( fig. 3k, l) Fig. 3c) shows a more curved dorsal margin and a more rounded posterodorsal projection than our otoliths of this species ( fig. 3d-f). As the specimen of G. couchi figured in Lombarte et al. (2018) comes from the same area (northern Adriatic Sea) and has a similar size like our specimens, the reason of this variation cannot be explained. The otolith of G. niger (Lombarte et al. 2018: Fig. 3i) displays a sharp posteroventral angle and a prominent preventral projection and thus differs from our otoliths of G. niger ( fig. 5g-i). Also the otolith of G. paganellus (Lombarte et al. 2018: Fig. 3e) differs from our material of G. paganellus ( fig. 4j-l) because the dorsal margin is rounded (vs. relatively flat in our otoliths) and the curvature of its posterodorsal projection is directed posteriorly (vs. dorsally). In both G. niger and G. paganellus, the differences could be due to ontogenetic variation because Lombarte et al.  (table 1). Moreover, also the otolith of G. cobitis (Lombarte et al. 2018: Fig. 3d) differs from our otoliths of this species ( fig. 4a-c) in its more rounded posteroventral margin (vs. angular), its smaller posterodorsal projection that is not facing dorsally, and in the relatively smooth margins (vs. crenulated). As the corresponding fish utilized in Lombarte et al. (2018) is within the size range of our specimens, the differences may again result from variation between geographically distant populations: Lombarte et al. (2018) sampled G. cobitis from the NW Mediterranean sea (NE of Barcelona, Catalan coast), while our G. cobitis specimens originate from the Adriatic Sea (Croatia and Montenegro). Overall, these comparisons confirm that the details of the otolith shape is generally species-specific in Gobius, but differences between geographically distant populations may occur and need further study. Nevertheless, we expect that the herein presented SEM images of the otoliths of 14 species of Gobius, together with the images shown in Lombarte et al. (2018), will provide a valuable source of information for the detection of extant species of Gobius in the fossil record. As a first step, we compared our dataset with some previously published fossil otoliths of which SEM images had been presented (Lin et al., 2015;Agiadi et al., 2013Agiadi et al., , 2018Agiadi et al., , 2019. On this basis, we found the interpretation of otoliths as G. aff. paganellus from the Upper Miocene (Tortonian) of northern Italy in Lin et al. (2015: Fig. 7: 8, 9) reliable because the fossil otoliths are similar to the extant ones, but differ in the rounded posteroventral angle (vs. angular) and the broader posterodorsal projection. Also the identification of Pleistocene otoliths as G. niger in Agiadi et al.
(2018: Fig. 5:68) and as G. vittatus in Agiadi et al. (2019: Fig. 4I, J) appears plausible; these fossil otoliths correspond largely with the otoliths of G. niger and G. vittatus of our data sets and that of Lombarte et al. (2018). On the other hand, the otolith from the Pleistocene that was classified as G. geniporus in Agiadi et al. (2019: Fig. 4F) is eroded and in our view not identifiable with certainty. Also the documentation of G. cf. paganellus from Pliocene sediments (Agiadi et al., 2013) and G. paganellus and G. cobitis from Pleistocene deposits (Agiadi et al., 2019) appear doubtful, as these otoliths clearly do not possess the typical shape of the respective species. However, based on the new comparative material presented here, they can be re-classified. The Pliocene otolith documented as G. cf. paganellus (Agiadi et al. 2013: Fig. 8:14) and the two otoliths from the Pleistocene named as G. cobitis (Agiadi et al., 2019: Fig. 4D, E) could well represent G. fallax, with which they share the general outline, inclination of the dorsal margin and blunt posterodorsal projection (see here fig. 3g, h); they would be the first fossil record of G. fallax. The two otoliths from the Pleistocene that were depicted as G. paganellus in Agiadi et al. (2019: Fig. 4G, H) are relatively high and have a conspicuously rounded anterodorsal margin, together with their pronounced and slightly pointed posterior projection they are very similar to G. incognitus and would be the first fossil record of this species.

Comparisons of the otolith morphology of Pomatoschistus with previous works
As said for Gobius, digital photographs or SEM images of extant species of Pomatoschistus are rarely provided in the literature and the otoliths of P. knerii ( fig. 6a-d), P. microps ( fig. 6i-l), and P. montenegrensis ( fig. 6n-p) are presented here for the first time. Photographs of an otolith of P. marmoratus and P. pictus were also depicted in Lombarte et al.
(2018: Fig. 1) and SEM images of P. marmoratus and P. quagga were shown in Gierl et al. (2018: Fig . 3); these species originated from the same region as our material, and their otolith morphology conforms well with our otoliths. Additionally, drawings of a series of otoliths of P. minutus were given in Nolf (2013: pl. 318), and also these images fit entirely with our otoliths of this species. Although the otoliths of a Pomatoschistus species can bear considerable intraspecific variability (see fig. 6), the here presented SEM images of their otoliths will greatly help to identify possible fossil representatives of extant species of Pomatoschistus in future studies.

Why can some goby species be more easily recognized than others?
Gobius auratus, G. geniporus, G. paganellus and G. vittatus could be separated by many otolith variables (table 8) and revealed also characteristic shapes based on the SEM images (figs. 3a-c, 4d-f, j-p). Gobius vittatus could also be separated from its congeners based on two body morphometric variables, but no body morphometric variable separated the other three species from one another. Previous studies have demonstrated clear links between otolith shape and environmental parameters (e.g., Volpedo & Echeverría, 2003;Lombarte et al., 2010;Vignon & Morat, 2010;Avigliano et al., 2016). Based on these works it can be hypothesised that the specific otolith morphometry of G. auratus, G. geniporus, G. paganellus and G. vittatus reflects particular habitat preferences, e.g., for a certain water depth or temperature or a specific type of substrate and diet. A review of the literature on the four species under discussion provides some support for such a relationship. (i) Herler et al. (2005) investigated the habitat use of G. auratus and G. fallax in the northern Adriatic Sea. The authors found that both species prefer water depths greater than 8 m, but that G. auratus clearly differed from G. fallax in its preference for more inclined substrates (>30-90°). (ii) Gobius geniporus inhabits solely sandy bottoms and does not consume algae; both traits are unusual for Gobius species (Zander, 2011). (iii) The rock goby G. paganellus is an intertidal-to-inshore, epibenthic species that lives under stones or on sheltered rocky shores with abundant algal cover (Miller, 1986;Zander, 2011). In their study on the composition of fish assemblages in six different macrohabitats in the Gulf of Trieste, Bonaca & Lipej (2005) found G. paganellus to be associated with a specific assemblage of algae (Wrangelia and Padina), while G. cobitis, G. cruentatus and G. fallax each occurred in all or almost all of the macrohabitats studied. It thus appears that G. paganellus has specific habitat preferences. (iv) The striped goby G. vittatus differs phenotypically from its congeners in having a broad, black, longitudinal band that extends along its head and body (Miller, 1986). It is found offshore on coralline substrates at depths of 5 to 34 (exceptionally 42) m, with a preference for moderately to steeply inclined bottoms composed of both rock and sand (Miller, 1986;Kovačić, 2007;Kovačić & Arko Pijevac, 2008). Its characteristic broad stripe may reflect an adaptation to some additional aspect(s) of its habitat. In the Pomatoschistus datasets, otolith morphometry was most significant for P. montenegrensis (table 9). In this case too, a relationship to environmental factors could be proposed, as P. montenegrensis is one of the few freshwater species in the genus Pomatoschistus (Šanda & Kovačić, 2009). Alternatively, phylogenetic factors may play a role as there is some evidence that P. montenegrensis actually might belong to a different genus, i.e. Ninnigobius (see Miller & Šanda, 2008;Freyhof, 2011;Thacker et al., 2019).
Regarding fossil otoliths, it is obvious that some otolith-based species of Gobius are easier to recognize than others. Examples are G. rostratus from the lower Miocene of the Mainz Basin, Germany (easily recognizable by its prolonged preventral projection; see Weiler, 1963: Figs. 190-198), Ponticola zosimovichi from the middle Miocene of Kazakhstan (well identifiable due to its strongly bulged anterodorsal margin; see Bratishko et al., 2015: Figs. 9-21), and several others. The results of our studies tentatively indicate that such distinctive otolith shapes may be related to specific environmental niches that these ancient species occupied. A further factor that could potentially influence otolith shape but cannot be evaluated yet is intraspecific communication based on acoustic signals. Experimental studies have revealed that gobies are capable of producing multiple sounds (Malavasi et al., 2008;Polgar et al., 2011;Parmentier et al., 2013;Horvatić et al., 2016), and acoustic courtship signals were demonstrated to be relevant for mating success in the painted goby Pomatoschistus pictus (Amorim et al., 2013). However, additional studies will be necessary before such a link between acoustic communication Abbreviations: *, **, *** indicates that one, two or three singular variables separated the species given in the columns from 78-89% of the congeners. AN(C)OVA means that ANOVA or, if covariance was detected, ANCOVA was used; LDA, Linear Discriminant Analysis. Table 8 Comparison of the species classification success in the reduced dataset of Gobius based on the different groups of variables (indicated by colors) and different statistical analyses. Bold numbers in rows indicate how many individual variables had separated the species given in the columns from ≥ 50% of the congeners (see table 2a2, d2, f2 for details). Percentages in rows represent the success of the respective multivariate analyses in separating the species given in the columns (100% means that the species is separated from all congeners by this variable; see table 2c2 and table 3 Table 9 Comparison of the species classification success in the reduced dataset of Pomatoschistus based on the different groups of variables (indicated by colors) and different statistical analyses. Bold numbers in rows indicate how many individual variables had separated the species given in the columns from ≥ 50% of the congeners (see table 4a2, d2, f2 for details). Percentages in rows represent the success of the respective multivariate analyses in separating the species given in the columns (100% means that the species is separated from all congeners by this variable; see table 4c2 and table 5  Abbreviations: *, **, *** indicates that one, two or three singular variables separated the species given in the columns from 78-89% of the congeners. AN(C)OVA means that ANOVA or, if covariance was detected, ANCOVA was used. and otolith morphology can be seriously considered (see Popper et al., 2005).

Conclusions
The outcome of this study significantly advances our knowledge of the interspecific differences between goby otoliths within a given genus. In agreement with previous work, we have shown that goby otoliths show a high species-specificity and are useful for taxonomic studies concerning this taxon. As the interspecific differences can be very delicate, the otolith shape analysis is a particular valuable tool for species identification. Studies of past goby species diversity using fossil otoliths may greatly benefit from the use of both singular otolith variables and elliptic Fourier shape analysis, but also traditional comparative morphology remains useful. Moreover, we uncovered certain body and fin proportions, and in the case of Pomatoschistus also meristic traits suitable to unravel species diversity when fossil skeletons are preserved. Notably, shared meristic characters among fossil gobies do not necessarily indicate that only a single species is present, as has been proposed in some previous works. The outcome of our study provides a new basis for the delimitation of fossil goby species in future studies, and will consequently help to better understand the evolution of goby diversity and biogeography through time.