Intra-specific variation and allometry of the skull of Late Cretaceous side-necked turtle Bauruemys elegans (Pleurodira, Podocnemididae) and how to deal with morphometric data in fossil vertebrates

Skull variation

The skull is one of the most variable structures in vertebrates because it concentrates several sensory organs, the brain, and the beginning of the respiratory and digestory systems, including chewing muscles (Smith, 1993). Consequently, the skull is the body part with more phenotypes used in vertebrate cladistic analysis (Rieppel, 1993), as seen in turtles, in which most cladistic analysis rely mainly on cranial characters (e.g., Gaffney, Meylan & Wyss, 1991; De La Fuente, 2003; Gaffney, Tong & Meylan, 2006; Gaffney et al., 2011; Joyce, 2007; Joyce & Lyson, 2010; Sterli et al., 2010; Sterli & De La Fuente, 2011; Anquetin, 2012; Rabi et al., 2013; Romano et al., 2014; Ferreira et al., 2015; Romano, 2016). Despite that, most skull materials found in paleontological record of turtles are rare and/or damaged due to the fossilization process bias, not allowing intraspecific comparisons or ontogenetic inferences on most fossil turtle species known. Some exceptions are found in Sánchez-Villagra & Winkler (2006) and Ferreira et al. (2015), who performed interspecific comparisons among fossil turtle taxa using skull material.

Preliminary analysis: caliper vs. ImageJ

Before carrying out others statistical analyses, we compared the same characters data set (Data S1) of a sub-sample by using two different approaches (=treatments): measurements taken using caliper and measurements taken using photographs via ImageJ. This comparison was necessary in order to evaluate whether or not the two measurements methods are significantly different. Then, we performed an One-way Analysis of Variance (ANOVA) comparing the 29 measurements in 12 specimens (LPRP0200, LPRP0369, LPRP0370, MN4322-V MN4324-V, MN6750-V, MN6783-V, MN6786-V, MN6787-V, MN6808-V, MN7017-V, and MN7071-V). Two groups of variables were established: measurements taken directly from specimens using caliper (preliminary data set 1) and the same characters taken from photographs of the same specimens using ImageJ (preliminary data set 2). All characters taken using photographs/ImageJ that did not show significant differences to their correspondents taken by caliper were used on the subsequent statistical analyses of form and shape differences among the sample of Bauruemys elegans. By doing that, the sample was increased without including error and incomparable characters (i.e., by using different measurement techniques).

We found most of the measurements do not differ statistically (p > 0.05) between the two treatments (caliper and ImageJ; Table 1). However, one measurement, length of maxilla (LMX), had statistical difference (p = 0.017) between the treatments, because the maxilla is a curved structure and thus the landmarks are in different positions (LM 24 is deeper and farther from the camera in relation to LM11) in relation to the plane the picture was taken. Given that no statistical differences were found in almost all characters, ImageJ could be an economic and time-saving tool for morphometric analyses from photographs (2D), and could be applied by scientists at distant institutions.

The study in situ of the material is preferable, although pictures are an economic alternative in where cases one is not able to handle the material. We must aware that one have to choose one of the two treatments to construct a morphometric matrix, otherwise it will be composed of values obtained by two diffent methods.

Univariate, multivariate and allometric analyses

Three analyses using the complete sample were carried out: (1) a descriptive statistics (mean, standard deviation, median, variance, maximum and minimum values) of all characters (Data S2), (2) an allometric analysis of length and width characters correlating them to total length and width measurements (Data S3), and (3) a multivariate non-parametric exploratory statistics via Principal Component Analysis (PCA). The latter was divided into two different PCAs: (3.1) using 27 characters from the raw data matrix (total length and width characters were excluded in this analysis; Data S4—because PCA is sensitive to large variations in the original measurements), and (3.2) using 27 characters that correspond to the proportions of each character from the raw data (i.e., original measurements) represented by its length or width characters divided by each individual total length or width (e.g., the length of MCZ4312 postorbital divided by the total length of the skull of this specimen; see complete data in Data S5). All statistical analyses were performed using the software PAST version 3.05 (Hammer, Harper & Ryan, 2001).

In the allometric analysis (analysis 2, Data S3), all characters were previously log-transformed and a linear regression was carried out separately for length and width characters, using the least square fitting approach for residuals. We established the allometries by considering the regression’s slope, i.e., the coefficient a, as following: positive allometry (a > 1), negative allometry (1 > a > 0), enantiometry (a < 0), and isometry (a = 1).

In the first PCA approach (3.1) we excluded total length and width characters because of their high influence on the PCA result, since higher values compose the majority of the summarized variance in PC’s (Mingoti, 2013), and because of the redundancy between these measurements and the others. We also assessed differences by applying two different substitution algorithms for missing data in PAST, using the default “mean value imputation” option (i.e., missing data are replaced by the column average), and the alternative “iterative imputation” option, which computes a regression upon an initial PCA until it converges to missing data estimations, replacing missing data by such estimations (Ilin & Raiko, 2010). The latter is recommended and, after comparing both results, we selected it (see Data S3 to visualize PCA results computed using PAST’s default option approach). The second PCA (3.2) was conducted to remove the effect of size (=growth) and perform an exploratory analysis of the shape alone. Six specimens were removed from this analysis because they were broken and the total length or width measures were not measurable.

The univariate analysis was made in order to quantify and describe the variation of the characters set in Bauruemys elegans skull, using the assumption that the sample is representative of a single population. The linear regression analyses allowed us to make inferences about osteological shape change related to size change, i.e., related to growth, by assuming that bigger specimens are older than smaller ones. This approach is, therefore, a study of allometry (sensu Huxley & Teissier, 1936; Huxley, 1950; Gould, 1966; Gould, 1979; Somers, 1989; Futuyma, 1993) and the assumption of correlation between size and aging is based on continuous growth to be common on extant turtles (Klinger & Musick, 1995; Shine & Iverson, 1995; Congdon et al., 2003). Since the use of a parametric statistic was infeasible due to the nature of the sample (i.e., a small dataset that do not show homoscedasticity and normality), the PCAs were used to search for a structure of the data that matches to the pattern found by Romano & Azevedo (2007) using shell characters (i.e., all individuals plotted inside the 95% confidence ellipse). If the pattern observed is similar to previous morphometric and taphonomic inferences, then the variation is not enough to assume that the sample represents different populations of Bauruemys elegans or a different species (see ‘Taxonomic considerations on the sample’). In other words, since a parametric test is not feasible with statistical confidence, the lack of structure in the PC plots were herein interpreted as a failure to falsify the single population hypothesis. All principal components were, therefore, analyzed but we present only those with higher variance.

Raw data

Shape characters (proportions)

Replacing missing data with mean values.

When applying “mean value imputation”, 53.99% of the variance were comprised by the first two principal components (PC1 = 35.29%; PC2 = 18.70%), both corresponding to shape, as all units of measurements were removed through the division of characters before carrying out the analysis. All specimens were comprised into the 95% confidence ellipse (Fig. 7A).

The first PC was positively related to the loadings values of LPA/TLS (L = 0.28), LMX/TLS (L = 0.38), LQU/TLS (L = 0.27), WPA/TWS (L = 0.23), SMX/TWS (L = 0.38), WMX/WTS (L = 0.35), WQJ/TWS (L = 0.48); the most negative values were LPO/TLS (L =  − 0.16) and WOP/TWS (L =  − 0.13). The second PC was positively related to LPA/TLS (L = 0.66), WPA/TWS (L = 0.32) WOP/TWS (L = 0.27), and negatively to LMX/TLS (L =  − 0.50) (see Table 4 for all loading values). It is interesting to note that most of highly-related proportions were in reference to bones associated either with feeding apparatus (squamosal, parietal, quadratojugal and jugal) or catching food and trituration surface (maxilla).

Table 4:

PCA loadings: proportion data.

Loading values of characters in the proportions data matrix related to the first two principal components in PCA, comparing the Mean Value (mv) approach with the Iterative Imputation (ii) approach.

Char.	Pc1 (mv)	Pc2 (mv)	Pc1 (ii)	Pc2 (ii)
LPF/TLS	0.003	−0.13	0.11	−0.30
LFR/TLS	0.001	−0.04	0.03	−0.02
LPA/TLS	0.28	0.66	−0.13	0.67
LVO/TLS	−0.002	0.05	−0.03	−0.02
LPAL/TLS	0.08	0.02	0.07	0.12
LPT/TLS	−0.05	−0.10	−0.02	−0.01
LBS/TLS	0.03	−0.17	0.11	−0.10
LBO/TLS	−0.02	−0.04	0.01	−0.04
LMX/TLS	0.38	−0.43	0.48	−0.18
LJU/TLS	0.16	0.01	0.16	0.14
LQJ/TLS	0.06	−0.09	0.21	−0.17
LQU/TLS	0.27	−0.07	0.28	0.05
LPO/TLS	−0.16	0.13	−0.18	0.03
LSQ/TLS	0.16	0.23	0.20	0.34
WPF/TWS	0.07	0.09	−0.001	0.11
WFR/TWS	0.07	0.13	0.02	0.05
WPA/TWS	0.23	0.32	0.08	0.33
SMX/TWS	0.38	−0.12	0.33	−0.01
WVO/TWS	−0.05	−0.04	−0.04	−0.10
WCO/TWS	−0.04	0.07	−0.11	0.04
WPAL/TWS	0.04	−0.07	0.04	−0.003
WBS/TWS	0.03	−0.05	0.02	−0.03
WMX/TWS	0.35	−0.05	0.30	0.03
WJU/TWS	0.18	0.01	0.26	0.19
WQJ/TWS	0.48	−0.003	0.41	0.20
WPO/TWS	0.02	0.01	−0.01	0.07
WOP/TWS	−0.13	0.27	−0.21	0.09

DOI: 10.7717/peerj.2890/table-4

Notes:

Char

characters

mv

Mean Value approach

ii

Iterative Imputation approach

The depositional context at the “Tartaruguito” site

The depositional environment at the Pirapozinho site is well-known from previous studies, which point out to seasonal floods in which turtles might have gathered in water bodies for foraging, followed by droughts that caused their death (Soares et al., 1980; Fulfaro & Perinotto, 1996; Fernandes & Coimbra, 2000; Henriques et al., 2002; Henriques et al., 2005; Suárez, 2002; Bertini et al., 2006; Henriques, 2006). This is, consequently, a case of several seasonal non-selective death events, with individuals representing semaphoronts connected temporally (between generations), thus comprising a single population (agreeing with Futuyma’s, 1993 population definition and used by Romano & Azevedo, 2007). We failed to disprove the null hypothesis that all individuals belong to a same population of Bauruemys elegans, agreeing with Romano & Azevedo (2007) conclusion using post-cranium data.

Taxonomic considerations on the sample

Many skulls sampled show taphonomic effects, such as cracks and crushing (Fig. 8). For instance, MN7071-V is notably the largest specimen of the sample and is represented in the uppermost positive side of the size-related PC2 axis (Fig. 6C). Although it is indeed a big specimen, it was clearly a taphonomic effect (crushing) that caused it to be larger than it really was (Fig. 8A). On the other hand, Bertini et al. (2006) indicated that turtle bodies have suffered little transportation or crushing in Tartaruguito site. We agree with this taphonomical interpretation of the site, as most specimens do not show huge breaks (Figs. 8B and 8D) that could cause misinterpretation of the morphometric results (the case of MN7071-V is an exception in our sample in this respect).

Another aspect is the presence of polymorphism in B. elegans. Romano (2008) presented an unusual carapace for the specimen MN7017-V, as having a seventh neural bone, differing from the diagnostic number of six neurals for this species, and with the diagnostic four-squared second neural bone not contacting first costals (Suárez, 1969b; Suárez, 1969c; Kischlat, 1994; Gaffney et al., 2011). The morphometric analysis performed by Romano (2008), using only shells, did not reveal significant statistical differences between MN 7017-V and other B. elegans specimens. We have included the MN7017-V skull in our analysis, and there was no variation to state anything apart from Romano’s (2008) conclusion that it is probably a polymorphic B. elegans specimen (Fig. 6C). Still, we reevaluated this skull and found the diagnostic characters for B. elegans. Therefore, all skulls included in our study belong to the same species (i.e., B. elegans).

Among the five valid fossil turtle species found throughout the Bauru Basin, only two have been collected at the Pirapozinho site so far (Romano et al., 2013). The first is B. elegans, which is recognized by both skull and shell materials; the second is Roxochelys wanderleyi Price, 1953 based only on shell material (Price, 1953; De Broin, 1991; Oliveira & Romano, 2007; Romano & Azevedo, 2007; Gaffney et al., 2011; Romano et al., 2013). So far, none R. wanderleyi with skull-shell associated body parts were collected, and thus we cannot claim that the skulls found at Tartaruguito site belong to this species until a skull-shell R. wanderleyi specimen be found, since all skulls analyzed here can be safely identified as belonging to B. elegans.

Bones of the upper temporal fossa and skull roofing

The temporal emargination of podocnemidid turtles is formed by the dorsal, horizontal plate of the parietal, the quadratojugal and the squamosal, with no contribution of the postorbital (Gaffney, 1979; Gaffney et al., 2011). This region (and bones) is associated with the origin of the adductor muscle fibers (m. adductor complex; Figs. 9A and 9B) (Schumacher, 1973; Werneburg, 2011; Werneburg, 2012; Jones et al., 2012; Werneburg, 2013), which run through cartilago transiliens of the processus trochlearis pterygoidei of the pterygoid and insert at the coronoid process of the lower jaw (Schumacher, 1973; Gaffney, 1975; Gaffney, 1979; Lemell, Beisser & Weisgram, 2000; Werneburg, 2011). These muscles promote the closure of the mouth, thus it is reasonable to associate the attachment surface to bite force and the latter to the prey hardness. Yet on the ventral flange of the squamosal originates the m. depressor mandibulae (Schumacher, 1973; Gaffney, Tong & Meylan, 2006; Werneburg, 2011; Fig. 9B), which causes the abduction (=opening) of the mandible.

The variation in this area of the skull in turtles was a matter of some studies (e.g., Dalrymple, 1977; Claude et al., 2004; Pfaller, Gignac & Erickson, 2011), which indicated allometric ontogenetic growth patterns of the bones in these regions. These authors were able to identify a high correlation with the increasing of muscle mass and shift in feeding features (Dalrymple, 1977; Pfaller et al., 2010; Pfaller, Gignac & Erickson, 2011). Moreover, there are changes in skull shape associated to the aquatic environment and foraging strategies, as suggested for emydid and testudinoid turtles by Claude et al. (2004). Although these studies focused on hide-necked turtles, the same morphoecological patterns can be applied to side-necked turtles, since there are habitat occupation similarities between side-necked and hide-necked turtles with implications to the skull morphology due to morphofunctional constraints (Schumacher, 1973; Lemell, Beisser & Weisgram, 2000), besides the adaptive selection regarding fresh water feeding strategies (see Lauder & Prendergast, 1992, Aerts, Van Damme & Herrel, 2001 and Van Damme & Aerts, 1997 for feeding strategies in freshwater turtles).

The high variance and positive allometric growth of the parietal (LPA: a = 0.38; WPA: a = 0.32), quadratojugal (LQJ: a = 0.16; WQJ: a =  − 0.06) and squamosal (LSQ: a = 0.30) lead to an increase in temporal emargination and, consequently, a greater area for attachment of the m. adductor mandibulae externus. The consequence of this would be the generation of large forces and high velocities during the fast closing phase of an aquatic feeder, as seen in Pelusios castaneus (Lemell, Beisser & Weisgram, 2000), and even a more powerful bite for crushing harder prey, as seen in Sternotherus minor (Pfaller, Gignac & Erickson, 2011). In addition, the lenghten of the squamosal would allow a greater insertion area of the m. depressor mandibulae and muscles of the hyobranchial apparatus (e.g., m. constrictor colli) (Schumacher, 1973; Gaffney, 1979; Claude et al., 2004; Gaffney et al., 2011; Werneburg, 2011). The m. depressor mandibulae is useful for an increased gape opening speed and the hyobranchial apparatus musculature is involved in backwards water flow generation by the lowering of the hyoid apparatus, two characteristics well reported for other pleurodire turtles as Chelodina longicollis, Chelus fimbriatus and Pelusios castaneus (e.g., Van Damme & Aerts, 1997; Aerts, Van Damme & Herrel, 2001; Lemell, Beisser & Weisgram, 2000; Lemell et al., 2002). Moreover, Claude et al. (2004) demonstrated that aquatic turtles with suction feeding mode possess longer skulls than terrestrial turtles, the squamosal being most prominent bone involved in this elongation and functionally related to the style of prey capture (=suction) as a support for mandible and hyoid muscles.

Also, Gaffney et al. (2011), in a comparison with other podocnemidid turtles, indicated B. elegans as having a “skull relatively wide and flat” (p. 12), which could be observed by the increasing of some bones, specially the postorbital (Figs. 3G and 4H), parietal (Figs. 3A and 3J), quadratojugal (Figs. 3I and 4F) and jugal (Figs. 3C and 5B). Comparing the postorbital allometry (better discussed below) with those of the bones in contact with it in the skull roof (frontal, parietal, jugal and quadratojugal; Gaffney et al., 2011), we observe an influence of the positive growth of the former into the others, leading to flattening and widening of the skull.

In a study assessing the bite performance in turtles, Herrel, O’Reilly & Richmond (2002) suggested that a higher skull is efficient in promoting stronger bite forces, specially in species which feed on hard prey, but they also pointed out that additions in bite forces may be achieved by “getting longer and larger” skull with no increasing in skull height. Thus, in addition to provide gains in muscle attachment area, by the growing of parietal, quadratojugal and squamosal, leading to a longer skull, a stronger bite and possibly a change in diet along the ontogeny. Also, the allometric growths of most of skull bones, particularly the positive allometry of the postorbital, suggests a more roofed skull in larger adults of B. elegans bigger adults. Given the allometric patterns aforementioned, B. elegans may have had a wide and flat but long skull, which would have compensated the loss of muscle volume and attachment area caused by widening and flattening the skull (Herrel, O’Reilly & Richmond, 2002). Correlations between a more emarginated skull and increases in the volume of the adductor muscles were also explored in a cranial evolutionary framework of stem-turtles by Sterli & De la Fuente (2010).

At last, Gaffney, Tong & Meylan (2006) and Gaffney et al. (2011) scored a character based upon the contact between quadratojugal and parietal bones (char. 13 of Gaffney, Tong & Meylan, 2006; char. 5 of Gaffney et al., 2011). They also state that this contact is present in Hamadachelys + Podocnemididae clade, with a large quadratojugal (state 1), in contrast to most of other Pelomedusoides (state 0: contact absent, as seen in Pelomedusidae, Araripemydidae and many bothremydids (e.g., Kurmademydini, Cearachelyini and Bothremydini); state 2: contact present with small quadratojugal in some Taphrosphyini, Bothremydidae). Indeed B. elegans possess a large quadratojugal, which means that the reduction of the postorbital evolved after Bauruemys node of divergence, as confirmed in performed cladistic analyses. However, we found a greater increasing (positive allometry) of the two measurements of the postorbital and this might have influenced the growth of parietal and quadratojugal, as well as the jugal (see below), so that the state 1 seen in B. elegans is possibly a consequence of allometric changes. This is easily seen when comparing the enatiometry of the width of the quadratojugal (WQJ: a =  − 0.06) and the slight increasing in the length of this bone (LQJ: a = 0.16) with the postorbital measurements. It also could have influenced the growth of the parietal, but to a lesser extent, as seen in the allometries of this bone (LPA: a = 0.38; WPA: a = 0.32).

When comparing the stem-Podocnemidinura species (i.e., Brasilemys, Hamadachelys) and stem-Podocnemididae (e.g., Bauruemys, Peiropemys, Pricemys and Lapparentemys), with the crown-Podocnemididae (i.e., Podocnemis lineage + Erymnochelys lineage) (Gaffney et al., 2011; Fig. 10), it is clear that an increasing in the parietal-quadratojugal contact has occurred along the podocnemidid lineage, and consequently led to a more roofed and less emarginated skull. We suggest that in B. elegans the small contact is due to the positive growth of the postorbital resulting in a more emarginated skull than other podocnemidids, as described by Gaffney et al. (2011). Yet within crown-Podocnemididae this bone suffered the opposite effect (i.e., small growth), showing variations in size and even being absent in some species (e.g., Podocnemis sextuberculata, Ruckes, 1937; Gaffney, 1979; Gaffney et al., 2011), though the emargination is still great. On the other hand, in the Erymnochelys lineage the postorbitals are large but the quadratojugal and parietal are large as well, leading to a greater contact between these bones and a well-roofed but less emarginated skull, being a reversion in Bairdemys venezuelensis and B. sanchezi within the Erymnochelys lineage (Gaffney et al., 2011). Therefore, the increase or decrease in the temporal emargination within Podocnemididae could be due to variation of allometric patterns in bones that form the skull roof, particularly the postorbital, quadratojugal, and parietal, among different lineages. Given that observation, we speculate that characters related to the form of the aforementioned bones (postorbital, squamosal, and parietal) are potencially more susceptible to homoplasy.

Bones of the lower temporal fossa

The lower adductor chamber in Pelomedusoides is formed externally and laterally by the jugal and quadratojugal, with the addition of the maxilla in some cases (e.g., Podocnemis spp. and Bairdemys sanchezi). The well developed cheek emargination, found in most but not all podocnemidid turtles (the exceptions are all within the Erymnochelys lineage but Bairdemys spp., Cordichelys antiqua and Latentemys plowdeni), is also part of the adductor chamber (Gaffney, 1979; Gaffney, Tong & Meylan, 2006; Gaffney et al., 2011). Internally and medially, the postorbital, the jugal, and the pterygoid compose the septum orbitotemporale, partially separating the fossa orbitalis from the fossa temporalis; along with the palatine, they aid to support the processus trochlearis pterygoidei of the pterygoid (Gaffney, 1975; Gaffney, 1979; Gaffney, Tong & Meylan, 2006). There is a passage medially to the process of the pterygoid and the septum orbitotemporale, running from the fossa orbitalis to the fossa temporalis, the sulcus palatinopterygoideus. The palatine and pterygoid form the floor of its passage, whereas the parietal, postorbital and frontal limit its upper portion. In this region, the m. adductor mandibulae fibers run through the processus trochlaris pterygoidei, and the m. adductor mandibulae internus (i.e., m. pterygoideus and pars pseudotemporalis; Fig. 9B) mostly originates throughout the pterygoid and parietal bones (Schumacher, 1973; Lemell, Beisser & Weisgram, 2000; Lemell et al., 2002; Werneburg, 2011). The m. adductor mandibulae internus fibers are involved in the jaw-closure system by generating counter forces (protraction) to the m. adductor mandibulae externus (retraction) (Schumacher, 1973; Lemell, Beisser & Weisgram, 2000; Lemell et al., 2002; Figs. 9C and 9D).

Variation of the upper temporal fossa has been studied in different turtles, such as various trionychids (Dalrymple, 1977) and Chelydra serpentina (Herrel, O’Reilly & Richmond, 2002). However, few studies report the variation of the lower adductor chamber, although both the upper and lower temporal fossa are anatomically and functionally coupled (Schumacher, 1973). Dalrymple (1977) identified a positive allometry in the width of the “temporal passageway” in trionychids. This area is related to the cryptodire pulley system (i.e., a processus trochlearis formed by the quadrate and opisthotic) and is analogous to the pleurodire pterygoid process, and thus can be comparable functionally (Gaffney, 1979). Herrel, O’Reilly & Richmond (2002) concluded that the increase of the bite force in turtles is due to either the increased height of the skull, leading to a more open angle of the processus trochlearis in relation to skull longitudinal axis, or to enlargement (in width and length) of the skull, because it allows more area for muscle attachment and volume. We observed the same pattern of growth change in B. elegans, as evidenced by the positive allometry of the parietal, postorbital, palatine and pterygoid bones. Other features were observed by Dalrymple (1977) in trionychids (e.g., height and width of the supraoccipital crest, lengthening of the squamosal crest and a development of a horizontal crest in the parietal) and were correlated to changes in skull shape with a shift in feeding habits, from softer to harder preys as individuals age. Again, this seems to be the case in B. elegans, as evidenced by the positive allometry of the squamosal and parietal bones.

The bones that mainly compose the skull rostrolaterally and the lateral emargination revealed a correlated allometric shape shift. Even so, jugal and maxilla showed small allometric variation (Figs. 4B, 4C, 6A, and 6B). The reduction of the jugal (WJU: a =  − 0.23) and quadratojugal (WQJ: a =  − 0.06) along with the small growth of the maxilla (WMX: a = 0.19) demonstrate a decrease in height at the anterior portion of the skull. Because of the contact between jugal and quadratojugal with the postorbital (and its increase; see previous topic), we suggest that the latter would possibly has affected the growth of the former bones. Moreover, the strong development of the postorbital would ultimately affect the width of the maxilla, which in turn would also affect the jugal. In contrast, the lengthen of this bone would be less affected (LMX: a = 0.39). In addition, there is a considerable increment in the stretch of maxilla (SMX: a = 0.70) (Fig. 3H) leading to a broader rostrum. Yet this could allow a greater area for crushing, as observed by Kischlat (1994) for B. elegans, but also related to ontogenetic growth (T Mariani, pers. obs., 2016). All these allometric changes indicate that B. elegans owns a more flattened and wider skull (Gaffney et al., 2011), which could have allowed greater bite forces generation (Herrel, O’Reilly & Richmond, 2002).

There are other morphological implications in which the lower adductor chamber bones are involved and that are worth discussing. As previously pointed out, three bones compose the septum orbitotemporale: pterygoid, jugal, and postorbital (Gaffney, 1979; Gaffney, Tong & Meylan, 2006). Together with the palatine, these three bones provide support for the processus trochlearis pterygoidei, where upon runs the tendon that connect the m. adductor externus complex into the lower jaw (Schumacher, 1973; Gaffney, 1975; Gaffney, 1979; Lemell, Beisser & Weisgram, 2000; Gaffney, Tong & Meylan, 2006; Werneburg, 2011). Nearby the process, many muscle fibers originate or cross towards their insertions points (Schumacher, 1973; Werneburg, 2011). The temporal emargination at the upper adductor chamber becomes more emarginted during growth. As a consequence, the attachment area for m. adductor mandibulae externus increase during aging, potentially generating stronger bite forces. The consequence of this temporal emargination indentation is that the trochlear process would must become more robust to support higher forces. We interpret that the positive allometries of pterygoid (LPT a = 1.37), postorbital (LPO a = 1.25 and WPO a = 1.36), and palatine (LPAL a = 1.11) could be a response to this robustness of the trochlerar process during growth. In other words, they would act together by giving more resistance to the area in which the high forces created by the m. adductor mandibulae externus are applied. Gaffney (1979) suggested this robustness occurs because muscle volume increase and, consequently, higher bite forces, so these three bones would reinforce the septum orbitotemporale in order to support and do not break when muscles are contracted. In addition to such reinforcement, the growth of palatine could be associated with a larger area for crushing preys such as mollusks and crustaceans, as pointed out by Kischlat (1994).

The m. adductor mandibulae internus and m. adductor mandibulae posterior (Fig. 9B), which originate at the quadrate, prootic, pterygoid, palatine, postorbital and the descending process of the parietal (Schumacher, 1973; Werneburg, 2011), are important during the jaw-closure phase. The importance of these muscles has been debated for early tetrapods with flat skull and aquatic lifestyle (e.g., Temnospondyli and Lepospondyli; Frazzetta, 1968), in which the internal muscle might have assumed the main function of closing the jaw (Werneburg, 2012). This also occurs in turtles with flat skulls and with poorly developed crista supraoccipitalis (e.g., Chelidae; Werneburg, 2011; Werneburg, 2012). However, B. elegans does not have a skull as flat as chelids, but has a long supraoccipital bone as well as a greater temporal emargination (Gaffney et al., 2011), indicating more area and volume available to m. adductor mandibulae externus (Dalrymple, 1977; Sterli & De la Fuente, 2010). The mechanical effects of adductor muscles upon the lower jaw during food capture has been demonstrated in some turtles (Schumacher, 1973; Lemell, Beisser & Weisgram, 2000; Lemell et al., 2002; Pfaller, Gignac & Erickson, 2011). These studies agree that besides acting to close the mouth, the m. adductor mandibulae internus executes counter protraction forces to the m. adductor mandibulae externus retraction forces, while m. adductors mandibulae posterior produce medial forces (Figs. 9C and 9D). The contraction of all these muscles together avoid displacements of the mandible and reduce stresses at the articulation (Schumacher, 1973; Lemell, Beisser & Weisgram, 2000; Lemell et al., 2002). The positive allometries of the bones of the lower adductor chamber of B. elegans, therefore, may reflect greater resistance for a more robust musculature of m. adductor mandibulae internus and m. adductor mandibulae posterior in response to higher forces created by external adductors. Besides, these muscles also play the main role in feeding, as proposed for aquatic feeders (Frazzetta, 1968; Werneburg, 2012), in addition to a larger area between the two tips of the maxilla (i.e., SMX a = 0.70) and a flattened skull.