A quantitative approach to determine the taxonomic identity and ontogeny of the pycnodontiform fish Pycnodus (Neopterygii, Actinopterygii) from the Eocene of Bolca Lagerstätte, Italy

Background The pycnodontiform fish Pycnodus is one of the representatives of the highly diverse actinopterygian fish fauna from the early Eocene Bolca Lagerstätte, representing one of the youngest and thus last occurrences of this extinct neopterygian clade. This genus has historically been used as a wastebasket taxon in regards to poorly known pycnodontiform fossils. Authors have argued over the specific status of the Bolca Lagerstätte Pycnodus in terms of how many species are contained within the genus with some arguing for multiple species and others suggesting lumping all Bolca specimens together into one species. Methods Here, we use a quantitative approach performing biometric and geometric morphometric analyses on 52 specimens of Pycnodus in order to determine if the morphological variability within the sample might be related to inter- or intraspecific variation. Results The analyses revealed that the variations of body shape, morphometric and meristic characters cannot be used to distinguish different morphotypes. On the contrary, our results show a remarkable link between shape and size, related to ontogeny. Discussion Differences in body shape of small (juvenile) and large (adult) individuals is probably related to different microhabitats occupation on the Bolca reef with juveniles sheltering within crevices on the reef and adults being more powerful swimmers that swim above the coral. Taxonomically, we suggest that the Bolca Pycnodus should be referred to strictly as Pycnodus apodus as this was the name given to the holotype. Additionally, an overview of species assigned to Pycnodus is given.


INTRODUCTION
Pycnodontiform fishes were a highly successful group of neopterygian fishes that colonized shallow marine, brackish, and freshwater habitats from the Norian to the A nearly complete specimen of P. lametae with crushed skull and missing caudal fin was reported from the freshwater Maastrichtian of Bhatali, India close to the Dongargaon area (Mohabey & Udhoji, 1996). However, the assignment of the name Pycnodus to this fish is dubious, since it lacks the post-parietal process typical of the Pycnodontidae (J.J. Cawley, 2018, personal observation). A more complete specimen of Pycnodus was found in the Palaeocene rocks of Palenque, Mexico (Alvarado-Ortega et al., 2015), which differs from the Eocene specimens from Bolca by having a greater number of ventral and post-cloacal ridge scales, less dorsal-and anal-fin pterygiophores and a large or regular-sized posterior-most neural spine. However, due to the inadequacy of the available sample, it is not possible to determine the actual differences between the Palaeocene material from Mexico and that from the Eocene of Bolca, and for this reason this taxon is referred to as Pycnodus sp.

Specimen sampling
We studied a selection of Pycnodus specimens from various museum collections, which were labeled either P. apodus, P. platessus, P. gibbus or Pycnodus sp. A total of 52 Pycnodus specimens from nine museum collections were used to obtain biometric information with 39 specimens from that sample being used for the geometric morphometric analysis as their higher quality preservation provided sufficient morphological information for the aim of this study (BM; Museo dei Fossili di Bolca; CM, Carnegie Museum, Pittsburgh, Pennsylvania; FMNH, Field Museum of Natural History, Chicago; MCSNV, Museo Civico di Storia Naturale di Verona; MGP-PD; Museo di Geologia e Paleontologia dell'Università di Padova; MNHN, Muséum National d'Histoire Naturelle, Paris; NHMUK, Natural History Museum of London; NHMW; Naturhistorisches Museum Wien; SNSB-BSPG, Staatliche Naturwissenshaftliche Sammlungen Bayerns-Bayerische Staatssammlung für Paläontologie und Geologie, München, Germany). For this analysis, the sample includes 17 specimens identified originally as Pycnodus sp., 14 specimens as P. platessus, six specimens as P. gibbus, and two specimens as P. apodus.

Geometric morphometric protocol
A total of 18 landmarks, four anchor points, and 10 semi-landmarks were digitized on photos taken from the studied specimens in the corresponding collections using the software TPSdig (Rohlf, 2005). Landmarks indicating homologous points were selected on the basis of their possible ecological or functional role following the scheme applied in some studies (Claverie & Wainwright, 2014;Tuset et al., 2014;Clarke, Lloyd & Friedman, 2016;Marramà, Garbelli & Carnevale, 2016a about shape variation in modern or extinct fishes (Fig. 1). The traits used match 12 out of 17 of the landmarks that was used for 57 species of Pycnodontiformes by . Additional traits used here are the anterior and posterior margins of the cloaca to see if they shift significantly between morphotypes; using four landmarks around the orbit instead of one in the center to capture more precisely the variability surrounding the orbit; not using the insertion of the pelvic fin as this character was rarely preserved in our specimens; the use of two landmarks for the cleithrum to capture variability in position and size of the pectoral fin instead of using just the one landmark for the insertion of the first pectoral fin ray due to the poor preservation of the pectoral fins in many specimens in contrast to the concave notch in the cleithrum.
The landmark coordinates were translated, rotated and scaled at unit centroid size by applying a Generalized Procrustes Analysis (GPA) to minimize the variation caused by size, orientation, location and rotation (Rohlf & Slice, 1990;Zelditch et al., 2004). The GPA was performed using the TPSrelw software package (Rohlf, 2003) and a principal component analysis was performed on Procrustes coordinates to obtain the Relative Warp (RW). Shape changes were shown along the axes using deformation grid plots. Missing values are replaced using the algorithm "Mean value imputation" (Hammer, Harper & Ryan, 2001). (2) ventral-most margin of orbit; (3) posterior-most margin of orbit; (4) anterior-most margin of orbit; (5) dorsal-most margin of orbit; (6) first dorsal pterygiophore; (7) last dorsal pterygiophore; (8) tip of dorsal lobe of caudal fin; (9) medial convex margin of caudal fin; (10) tip of ventral lobe of caudal fin; (11) final anal pterygiophore; (12) first anal pterygiophore; (13) posterior cloacal scale; (14) anterior cloacal scale; (15) joint between quadrate and prearticular; (16) ventral-most concave margin of cleithrum accommodating pectoral fin; (17) dorsal-most concave margin of cleithrum accommodating pectoral fin; (18) Point of contact between neurocranium and vertebral column. Red circles marked with an asterisk are anchor points for the semi-landmarks. The semi-landmarks are represented by small white circles and are split into two sets; the first set consists of seven semi-landmarks between the tip of the dermosupraoccipital and the base of the first principal caudal fin ray; the second set has an additional seven semi-landmarks between the base of the ventral-most principal caudal fin ray and the antero-ventral corner of the cleithrum. Photo credit: Jürgen Kriwet.
Full-size  DOI: 10.7717/peerj.4809/ fig-1 Two non-parametric tests were performed to analyze the quantitative morphospace occupation of our Pycnodus specimens. In order to assess the degree of overlap between morphospaces, an analysis of similarities (ANOSIM) (Clarke, 1993) was performed on the entire dataset of standardized morphometric and meristic parameters. PERMANOVA (Anderson, 2001) was used to test similarities of in-group centroid position between the different groups representing a species of Pycnodus. Euclidean distances are the distance measure chosen for both tests. All statistical analyses were performed in PAST 3.18 (Hammer, Harper & Ryan, 2001).
Since the studied specimens vary greatly in size (smallest being 4.0 cm and largest being 30.6 cm) we also investigated whether size could be correlated with shape change in Pycnodus and enable us to see whether and how body shape changes throughout ontogeny. To analyze the relationship between size and shape, we performed a partial least square analysis (PLS) using the software TPSpls (Rohlf & Corti, 2000). Alpha (level of significance) was set to 0.05.

Biometric analyses
We used 11 meristic counts (number of vertebrae, ribs, scale bars, paired fin rays, median fin rays, median fin pterygiophores, caudal fin rays, and arcocentra interdigitations) and 19 measurements (see Supplementary Material) in order to capture morphological variability, to test the homogeneity of the sample, and confirming its assignment to a single morphotype. Histograms were used to illustrate the variation of morphometric and meristic data in order to ascertain if more than one morphotype of Pycnodus could be identified. Histograms can be problematic in accurately capturing the distribution of data (Salgado-Ugarte et al., 2000) so we also used Kernel density estimators to determine the presence of a normal (Gaussian) distribution of the meristic data. Least squares regression was used to obtain the relationship between standard length (SL) and all other morphometric variables. Specimens of possible additional taxa were indicated by the presence of statistical outliers from the regression line (Simon et al., 2010) and will require additional scrutiny in order to truly differentiate the outlier from all other specimens. The linear regression results were shown using scatterplots. Log-transformed data were used to perform the least squares regression in order to determine the degree of correlation between the SL and all other morphometric variables.

Geometric morphometrics
The RW analysis produced 38 RWs with the first three axes together explaining about 73% of the total variation. Figures 2 and 3 show that there is significant overlap between the morphospaces of the Pycnodus taxonomic groups and the thin plate splines show the changes in shape along the axes. Negative values on RW1 (56.3% explained) are related to Pycnodus specimens with large orbits and deep bodies while positive scores identify Pycnodus with reduced orbits and elongated bodies. Negative values of RW2 (10.3% explained) show specimens having the pectoral fin with a wide base moved higher up the body alongside a long caudal peduncle (Fig. 2). Conversely, on positive scores of RW2 lie specimens with pectoral fin with a narrower base located more ventrally on the body alongside a small caudal peduncle. The negative values of RW3 (5.9% explained) show the skull becoming deeper and more elongated with the dermosupraoccipital in particular reaching far back (Fig. 3). The body becomes shallower near the caudal peduncle with the cloaca shifting posteriorly, as well as the dorsal apex. Positive scores of RW3 are related to a shorter and shallower skull with the body becoming deeper close to the caudal peduncle and the anterior shift in the cloaca with the body becoming deeper just anterior to the cloaca. The dorsal apex shifts forward in position.
Analysis of similarities performed on the first three axes suggests that there is strong overlap between groups, showing they are barely distinguishable from each other (r-value is 0.10 and p > 0.05; see Table 1), except for a single pairwise comparison between Pycnodus sp. and P. platessus (p < 0.05). The PERMANOVA suggests the same trend ( Table 2), showing that group centroids are not significantly different on each pairwise comparison (f-value is 2.83), except between Pycnodus sp. and P. platessus (p < 0.05) which lends significance to the overall p-value (<0.05). Significant differences detected between Pycnodus sp. and P. platessus can be explained with the fact that the indeterminate Pycnodus specimens show a wide range of morphologies, with the extreme shapes ranging from negative to positive values of all the first three axes.  The PLS performed on the entire sample ( Fig. 4) revealed a strong and significant correlations between size and shape (r = 0.88; p < 0.05), therefore suggesting that different shapes of the individuals are related to changes in shape of different ontogenetic stages. Small-sized individuals are associated with larger orbits, deeper skull and body shape, long skull, higher position of pectoral fin and a wide, indistinct caudal peduncle that is in distant proximity to both medial fins. Larger individuals, on the other hand, have a reduced orbit, shallower skull and body depth, shorter skull, lower position of pectoral fin and narrow caudal peduncle in close proximity to both medial fins. The PLS analysis therefore suggests that the morphological variations of the orbit, body depth and caudal peduncle are strongly related to ontogeny. Morphometrics and meristic counts for all the studied specimens are given in Tables 3  and 4, respectively and mean biometric parameters are given in Table 5. Most of the histograms based on meristic counts (Fig. 5)
regression performed on morphometric characters (Fig. 6) shows that all specimens fit within the cloud of points near the regression line and that no particular specimens of Pycnodus deviates from this line. Variation in meristic values and the few outliers in partial least square regression analyses have been interpreted here as measurement errors due to incomplete preservation of some structures due to taphonomy or incomplete mineralization in juvenile individuals. The high values of the coefficient of determination (r 2 ) ranging from 0.76 to 0.99 (Table 6) indicate a high degree of positive correlation between SL and each morphometric character. Linear regression analysis also revealed the highly significant relationship between the SL and all morphometric characters (p < 0.001). Neither morphometric nor meristic characters are therefore useful in determining two or more different morphologically identifiable species within Pycnodus, strongly supporting Blot's (1987) hypothesis that only one species (P. apodus; see also below) is present in the Bolca Lagerstätte.

Intraspecific variation of Pycnodus apodus
The results demonstrate that all Pycnodus species cannot be separated morphologically using the morphometric traits used herein in a quantitative approach, supporting the intraspecific variation hypothesis of Blot (1987). P. gibbus is a problematic taxon to identify due to Heckel (1856) not mentioning exactly which specimen he used to designate the specific name for P. gibbus. Blot (1987) mentions that Heckel worked on specimens from the NHMW in order to erect P. gibbus. However, such specimens could not be found and so the holotype still remains unknown. However, Heckel (1856, plate 8) does illustrate a specimen of P. gibbus and it conforms with what we have found to be the juvenile morphotype in our sample lending credence to the hypothesis by Agassiz (1833-1844) that the specimens he studied were specifically the juvenile of P. platessus. One of the characters separating P. gibbus from P. platessus (Heckel, 1856, plate 8, Fig. 4) is the number of interdigitations between vertebrae (P. gibbus: two; P. platessus: three-four). However, a survey of the vertebral column of all our specimens reveals two to be the predominant number of interdigitations, including specimens labeled P. platessus and P. apodus. Apart from specimens where the degree of preservation was insufficient to do a count, only one specimen (MGP-PD 8868C) has three interdigitations which we ascertain to be due to intraspecific variation. Blot (1987 , Table 6) also did not see any difference in the number of interdigitations between P. gibbus and P. platessus. As suggested by Grande & Young (2004), ontogenetic variation of morphological characters actually represents a primary source of intraspecific variation; this is confirmed by our analysis, specifically by the morphological changes mostly occurring along RW1 in the morphospace that are related to ontogeny and the very significant results deriving from the PLS analysis. The unimodal (Gaussian) distribution cannot be seen in most of the meristic data, as revealed by the Kernel density estimator on the frequency histograms (Fig. 5), due to the fact that the sample is too small to detect high frequency of mean values (De Baets, Klug & Monnet, 2013). However, a few meristic characters reveal a domination of intermediate values and comparably rare extremes, which is typical of a homogenous population. Furthermore, the linear regression showed a significant dependence between SL and all morphometric variables, therefore suggesting that morphometric characters are not useful to distinguish different taxa. Meristic and morphometric data seem to show that all specimens studied belong to a single taxonomic entity (see Dagys, Bucher & Weitschat, 1999;Dagys, 2001;Weitschat, 2008;Marramà & Carnevale, 2015a;Sferco, López-Arbarello & Báez, 2015). shows some notable differences between the juvenile and larger specimens including the degree of ossification, particularly in the skull and caudal fin, being reduced in juvenile in comparison to adults and the notochord not being surrounded by arcocentra in juveniles whereas it is completely enclosed in adults. The so-called gibbosity that Heckel (1856) used to distinguish P. gibbus from P. platessus is formed by the angle of the anterior profile and the axis of the body. This angle decreases in larger specimens of Pycnodus from 70 to 55 (Blot, 1987) due to the skull roof moving posteriorly during growth revealing that this character probably does not denote a species but a growth stage within a single species. The high vertebrae length/body depth ratio said to be another indicator of P. gibbus is something that also decreases during growth. When Blot plotted all Pycnodus specimens onto a growth curve (Blot, 1987, fig. 32) P. gibbus fitted into the curve neatly on the lower end of the growth curve.

Habitat use during ontogeny
Our morphometric results show that the morphology of the smaller individuals differ significantly from that of the adults and that Pycnodus, like extant actinopterygians, would go through morphological changes throughout ontogeny. Large eye size found in the smaller Pycnodus specimens is usually a sign of the specimen being in a juvenile stage as can be seen in many extant teleosts (Pankhurst & Montgomery, 1990). Large eye size in pycnodonts has been related to behavioral flexibility and possible nocturnal behavior (Goatley, Bellwood & Bellwood, 2010). This could also apply for the Bolca Pycnodus although the individuals with the largest eyes (juveniles) are not believed to be more nocturnal as larger eye size in smaller fishes is a natural consequence of ontogeny. The deep body shape of the smaller Pycnodus specimens can be interpreted as a sign that the juveniles live within the branches of corals and as they get bigger they start to occupy the water column above the reef. Coral reefs composed of scleractinian coral colonies have been reported in situ (Vescogni et al., 2016) and were probably even more extensive based on abundant remains from the laminated and massive fossiliferous limestone from Pesciara and Monte Postale sites. This change to a benthopelagic lifestyle is also supported by the more fusiform body and the narrower caudal peduncle (Webb, 1982) seen in larger specimens. Ecologically similar extant analogues to Pycnodus, fishes of the genus Lethrinus undergo ontogenetic changes in head shape as they grow in size but their body depth in relation to length does not change drastically during growth (Wilson, 1998). The sparid species Diplodus sargus and D. puntazzo also spend their time as juveniles in crevices in the rocks in shallow water 0-2 m deep and move to rocky bottoms and sea grass beds when adult (Macpherson, 1998). However, their ontogenetic trajectory differs from Pycnodus as they are more elongate as juveniles and body depth increases with age. Juvenile carangids also have a deeper body than that seen in adults (Leis et al., 2006) and are found within lagoonal patch reefs (Wetherbee et al., 2004) only moving out of this habitat when larger than 40 cm and becoming more pelagic in their habitat preferences (Kuiter, 1993;Myers, 1999). Eurasian perch (Perca fluviatilis) go through three different feeding modes during their life span; zooplanktivory, benthic macroinvertebrate feeding, and piscivory. The middle stage, benthic feeding results in them shifting to the littoral zone where they have a deeper body and longer fins which aid in maneuverability whereas piscivores and zooplanktivores have a similar body type due to both life stages living in the pelagic realm (Hjelm, Persson & Christensen, 2000).
Ontogenetically-related habitat changes also occur in other coral fishes, such as labrids, in which the pectoral fins increase their aspect ratio as these fishes grow in size, enabling them to increase their use of the water column while juveniles stay closer to the bottom (Fulton, Bellwood & Wainwright, 2001). Since both juveniles and adults of Pycnodus are found in the Bolca Lagerstätte, we hypothesize that unlike many modern coral reef fishes, which significantly change the habitat during ontogeny (Nagelkerken et al., 2002;Dorenbosch et al., 2005aDorenbosch et al., , 2005bAdams et al., 2006;Nagelkerken, 2007;Nakamura et al., 2008;Shibuno et al., 2008;Kimirei et al., 2011), there is a shift instead in microhabitat use within the reef, in this case juveniles living within coral crevices to adults roaming over the coral reefs.

CONCLUSION
The quantitative approach here performed supports the hypothesis of Blot (1987) that the various P. nominal species (P. apodus, P. platessus, P. gibbus) from the Eocene Bolca Konservat-Lagerstätte actually belong to a single species. Due to the holotype of Pycnodus being given the specific name of apoda, all known specimens of Pycnodus from Bolca should be referred to as P. apodus. Most of the morphological variation can be explained by the close correlation between morphometric changes and ontogeny, with juveniles and adults occupying different parts of the morphospace. The morphometric differences between juveniles and adults may be due to occupation of different habitats with juveniles sheltering among cover and adults being better adapted to a roaming lifestyle swimming over the benthos to feed. The complex taxonomic history shows that most species typically referred to as Pycnodus are different taxa altogether (e.g. all Jurassic and Cretaceous Pycnodus specimens) and with the majority of Palaeogene Pycnodus being represented by isolated dentition it seems that the only definitive articulated skeletal remains attributed to the genus Pycnodus are P. apodus from the Bolca Lagerstätte and Pycnodus sp. from south-eastern Mexico (Alvarado-Ortega et al., 2015). Future studies should analyze other problematic pycnodontiform taxa such as the widely distributed Gyrodus from the Middle Jurassic to the Early Cretaceous (Kriwet & Schmitz, 2005) to investigate if intraspecific variation might partially explain the supposed diversity of species this genus contains.