Neck shape was documented for different positions of neck motion (Fig. 2A–B). Therefore, photographs of manually articulated necks as well as necks in life condition were compared. In addition to illustrations from older literature, I used the recent data of Werneburg et al. (2015a), who made medical CT-scans and fluoroscopy of neck motion in living specimens (Table 1). In total, the neck mobility of 39 turtle and two squamate species (as outgroup) were analyzed (Table 1, S4) by measuring angles between adjacent vertebrae. Werneburg et al. (2015a) also calculated and simulated life mobility in the fossil stem turtle Proganochelys quenstedti (Staatliches Museum für Naturkunde Stuttgart; SMNS 16980). The neck positions of this species, as reconstructed by those authors, were used herein to infer the ancestral shape of neck curves.

The following neck positions were documented for each species separately: the maximal (A) dorsal, (B) ventral, and (C) lateral neck flexion (i.e., bending of the neck); (D) the maximal extended neck position; (E) the neutral neck position (i.e., full overlap of articular facets of the adjacent zygapophyses); and (F) the neck positions during maximal retraction (Fig. 2A–B). Using different imaging software (Amira 5.2.2., Meshlab v1.3.0, IrfanView 4.32), I generated screen shot pictures of those neck positions. All necks at lateral flexion and the retracted necks of the stem turtle and pleurodires were photographed in a ventral planar view of the necks. All other neck positions were photographed in a lateral planar view (Fig. 2B).

Tps-files, which refer to image folders, were created using the software TpsUtil. Using TpsDig2, the most ventral and medial points in the centrals’ articulations were defined as homologous landmarks in fossil and extant material (Fig. 2Bc–i, Table 2). Each neck motion was measured separately in each species. In extant specimens and in the fossil P. quenstedti, the articulation between skull and neck and the anteroventral most point of the upper jaw were scored correspondingly (Fig. 2Ba–b). For other fossils, no skulls were available and only the landmarks between the vertebrae were measured (Fig. 2Bc–i). The various pieces of software used are available at http://life.bio.sunysb.edu/morph/.

Using the software PAST 2.15 (Hammer et al., 2001), 2D-geometric morphometric analyses were performed to characterize the shapes of the neck/head curves. To remove translation, rotation, and size effects, the Generalized Procrustes method (Rohlf and Slice, 1990) was used, which normalizes the coordinates of the landmark data with the help of a calculated centroid. Principal component analyses (PCA) were performed (Fig. 3 and Fig. 4, S1–16). Eigen values and percent variance of each analysis (Table 3) as well as PC1 scores and the normalized size of each curve (Table 4) were documented.

For each species, the relative sizes of emargination (depth divided by breadth) were documented and named as “emargination quotients” (EQs) (Table 1, Fig. 5). Depths and breadths of the emarginations were measured in planar lateral skull view for anteroventral emargination (Fig. 5A) and in planar dorsal skull view for posterodorsal emargination (cf., Fig. 5B).

The EQs describe the relative expansion of the respective emarginations. The length of the anteroventral emargination is measured between the posteroventralmost point of the upper jaw and, following a straight line, the ventralmost point of the ear capsule. From the middle point of this line, a straight line was measured to the deepest overall expansion of the emargination (Fig. 5A). For the posterodorsal emargination, a straight line is measured between the posteriormost point of the supraoccipital crest and the posteriormost point of the squamosal crest (cf. Fig. 5B). From the middle point of this line, the depth of the emargination is measured towards the deepest most overall expansion of this emargination.

For each species, EQs were compared to the neck curve shapes. For that, the EQs of anteroventral and posterodorsal emargination (Table 1) were correlated to the scores of PC1 (= largest shape variance; Table 4) of the respective neck curves (Table 5A–B). Centroid size of each curve was also compared to the EQs to estimate size correlations (Table 5C–D). Phylogenetic effect was corrected using phylogenetic independent contrast (Felsenstein, 1985) using the PDAP package in Mesquite 2.75 (Maddison and Maddison, 2011). A time scaled composite phylogeny of turtles was used as template (Fig. 6). Interrelationship of extinct taxa follows Werneburg et al. (2015b), interrelationship of extant taxa follows Thomson and Shaffer (2010), but the position of Platysternon megacephalum is that of Joyce et al. (2013). Divergence times of major turtle taxa follow Joyce et al. (2013) with adaptations by Werneburg et al. (2015b: caption of Table S9). Calibration numbers in million years (my) were rounded off to the nearest integer. For a non-calibrated node between two nodes of a known age, the branch lengths between the known ages were equally separated. Divergence times of specimens of the same species were defined as 1000 years. Specimens that were treated with different methods to detect neck shape were defined as terminal taxa and set into a polytomy (number of degrees of freedom to subtract for soft polytomies was set to 0).

Anatomical terminology and abbreviations mainly follow Werneburg, 2011, Werneburg, 2012, Werneburg, 2013a and Werneburg, 2013b. av = anteroventral emargination (orange), b = breadth of an emargination in planar view, cnubra-I and -II = cornu branchiale I and II, cpuhyo = corpus hyoideus, CV1/-2 = cervical vertebra 1 (atlas) and 2 (axis), cstsprocc = crista supraoccipitalis, cstsqu = crista squamosalis, d = depth of an emargination in planar view, EQav/pd = emargination quotient for anteroventral/posterodorsal skull emargination [i.e. depth divided by breadth of the emargination (see Fig. 5A)], itf = infratemporal fenestrum, occ = occipital region, pd = posterodorsal emargination (blue), stf = supratemporal fenestrum/cleft, tym = cavum tympanicum, V = ventral neck flexion.

Specimens are listed in Table 1. AMNH = American Museum of Natural History, New York, USA; FMNH = Field Museum of Natural History, Chicago, USA; IW = personal specimen ID of I.W.; SMF = Senckenberg Forschungsinstitut und Naturmuseum Frankfurt, Frankfurt am Main, Germany; TCT = Teaching collection of Geowissenschaftliches Institut Universität, Tübingen, Germany.

The morphospace occupied by the neck flexion in various planes (dorsal, ventral, lateral) and the neutral and extended neck curves are largely separated from the morphospace occupied by the neck retraction curves (Fig. 3).

Pleurodires have two types of lateral neck motion: a U-shaped lateral neck flexion and an S-shaped neck retraction. Both strongly differ in the morphospace that they occupy (Fig. 3 and Fig. 4). The cryptodiran and pleurodiran retraction curves are more similar to each other than equivalent neck flexions (i.e., simple bends in either of the four possible directions), but this is an artefact created by plotting the neck motion onto a two-dimensional plot, because neck retractions take place in contrary directions in these two groups of turtles (Fig. 1B–C).

The morphospace that is occupied by the curves of extended necks shows a strong overlap with that of the neutral position; however, the latter is more similar to the morphospace occupied by the curves of ventral flexion.

During dorsal and ventral flexion, the neck is limited by the carapace and plastron (cf. Fig. 2A), whereas the amount of lateral flexion is restricted by the limbs. Both cascades are reflected in the outlines of the respective curves (Fig. S7, S9).

Curve shapes of neck flexions clearly differ between stem fossils, pleurodires, and cryptodires and all turtles are clearly separated from the outgroup species (Figs. S7–12).

When plotting neck curves of lateral flexion, dorsal flexion, and all modes of retraction into one diagram, the following pattern is visible (Fig. 4). The lateral neck flexion of the outgroup (squamates) is more similar to the cryptodiran than to the pleurodiran mode of lateral flexion. Pleurodire retraction curves plot within the morphospace that is occupied by the retraction curves of cryptodires, and do not overlap with the pleurodiran lateral flexion curves.

The neck curves of cryptodires occupy a much larger morphospace than those of pleurodires (Fig. 3 and Fig. 4), which may be interpreted as an artefact of the fewer pleurodiran species studied herein (eight pleurodire vs. 26 cryptodire species). However, given the generally larger ecological diversity of cryptodires (Ernst and Barbour, 1992), the morphospace distribution may represent natural conditions.

The morphospace occupied by the neck curves of P. quenstedti is limited to a restricted area reflecting the ancestral low mobility of its neck (Fig. 4). The retraction curve of P. quenstedti (PqR in Fig. 4) plots within cryptodiran and near the pleurodiran dorsal flexion and plots nearer to the lateral flexion (PqL) than to dorsal flexion (PqD) of the same species (Fig. 4).

The minimal span tree (minimal shape difference; gray lines between points in Fig. 4) shows a close similarity between the cryptodiran retraction curves and the dorsal flexion curves.

The neck curve analyses were performed with two data sets, with seven landmarks (Fig. 2Bc–i) along the cervical column and with all ten landmarks (Fig. 2Ba–j), including the skull (exemplified in Fig. S15). Although the 7-landmark diagrams show a clearer picture, the general interpretation of the 10-landmark diagrams, with less data (no skull in most fossils), remains the same. Divergences occur due to the certain independent position and flexible orientation of the head during all neck motions (cf. Fig. 1B). The flexibility between the head and the neck is permitted mainly by the articulation between the skull and the first cervical vertebra (George and Shah, 1955a, George and Shah, 1955b, Ogushi, 1913, Shah, 1963 and Werneburg, 2011). Its flexibility does not influence the general distribution of neck curve shapes in morphospace.

By comparing shape (PC1) and centroid size (CS) of the neck curves (Table 4) with the emargination quotients (Table 1, Fig. 5C), the following correlations were revealed (see Table 5). There is significant correlation between the centroid size of the cryptodiran retracted neck and the posterodorsal emargination (r = 0.6645, P = 0.0007). Also PC1 of the cryptodiran retracted neck (r = 0.751, P = 0.0778) and ventral neck flexion in all turtles (r = 0.5065, P = 0.1641) display a nearly statistically significant correlation with the expansion of posterodorsal emargination. For all other traits, the correlations are not statistically significant. That also includes all tested correlations between the neck parameters and the expansion of the anteroventral emargination. Significant correlation, however, exists between the expansions of both anteroventral and posterodorsal emarginations (r = 0.388, P = 0.008).

The analyses of curve shape disparity show a large range in the mobility of lateral and dorsal neck flexions and neck retractions in turtles. The ventral flexion, however, shows less disparity (Fig. 3). This corresponds to the detected differences in raw joint mobility detected by Werneburg et al. (2015a). Apparently, motion of the head to the lateral side during retraction and when searching for food to the left and right side requires more flexibility of the neck than ventral movement. The same seems to be the case for dorsal motion when snorkeling and food catching. Disparity in ventral flexion, however, appears to be less important given the flat association of the animals with the bottom soil, which is facilitated by the form of the flat plastron. Moreover, the presence of the plastron limits the range of ventral neck flexibility. This is also true for aquatic turtles, which, in theory, could largely move their neck ventrally [illustrated by Werneburg et al. (2015a)] when swimming in the water column. Particularly, in the aquatic pleurodires, the specific anatomy of the neck vertebrae, which are strongly adapted to lateral retraction (Herrel et al., 2008, Werneburg et al., 2015a, Werneburg et al., 2015b and Williams, 1950), appears to prevent much of the ventral motion. Moreover, an excessive ventral flexion of the neck of all pleurodires and cryptodires alike would result in a strangulation of the trachea at the plastron. This appears to be the main reason for the limited ventral flexibility of the neck, because in theory, the neck allows much more ventral flexion than is performed in life (illustrated by Werneburg et al., 2015a).

One major problem of reconstructing the anatomy of the vertebral column in fossil vertebrates is the definition of the neutral, namely the resting position of the neck (Christian, 2002, Christian and Heinrich, 1998 and Taylor et al., 2009). Werneburg et al. (2015a) have demonstrated that the neutral position of the turtle neck is established when the articular facets of the adjacent vertebrae fully overlap. The present study confirms that finding. The shape distribution of the extended necks, with their more or less straight orientation, occupies a different morphospace when compared to the neck curves of the neutral neck position. But it is worth mentioning that there is still much overlap between the curve lines of neutral and extended necks (Fig. 3, light and dark gray). As such, when reconstructing a more or less straight vertebral column in a fossil model, there is a certain chance that the actual neutral position could be reconstructed. However, to accurately reconstruct the resting position, as mentioned above, the articular facets of the zygapophyses should show the greatest overlap.

The origin of the sideward, pleurodiran retraction mode has been illustrated to be derived from the ancestral type of neck retraction as reconstructed for P. quenstedti (Fig. 1C to B). To enable the retraction of the elongated pleurodiran neck, only a middle kink had to be evolved, which involved zygapophyseal elongation and other modifications of the related vertebrae, particularly in cervical five and six (Van Damme et al., 1995, Weisgram and Splechtna, 1990, Werneburg et al., 2015a and Werneburg et al., 2015b). Based on similarities of angular interrelationship along the cervical vertebral column, the vertical, cryptodiran mode of neck retraction has been hypothesized to be derived from the dorsal neck flexion found in stem turtles (Werneburg et al., 2015a). Given the non-specialized anatomy of the neck vertebrae in stem fossil taxa (Werneburg et al., 2015b) and the ability of intervertebral rotation between the cervicals that the authors detected, one could hypothesize that the cryptodiran mode of retraction could also, like the pleurodiran mode, be derived from the ancestral mode of retraction (Fig. 1C to A). In actual fact, the curve shape of the retracted neck in P. quenstedti is more similar to modern cryptodiran retraction than the dorsal neck flexion of P. quenstedti is. Namely, the dorsally flexed neck curve of P. quenstedti is more similar to the plesiomorphic neck mobility of the outgroup (Fig. 4). That result shows that not only the angular interrelationship, but also the overall shape of a neck curve should be taken into account to reconstruct the evolution of neck mobility.

The eighth, last cervical vertebra, which articulates with the body, enables the major angular rotation of the neck during retraction in modern turtles. Whereas cryptodires show comprehensive modifications of the eighth cervical vertebra (Dalrymple, 1979), the cervical eight of pleurodires does not show much change when compared to the ancestral condition (Werneburg et al., 2015b and Williams, 1950). Relative to the retracted neck in stem turtles (Fig. 1A; Werneburg et al., 2015a), an additional loop was acquired in pleurodires. In those, one half of the neck is laid towards one side and the other half towards the opposite side of the body. The articulation between cervical eight and the body did not have to change its anatomy because cervical eight simply swings only to the other lateral body side, hence flexibility of this articulation did not change compared to the ancestral condition. In cryptodires, a modification of the eighth cervical evolved to enable the articulation of neck and body towards a completely different, namely a vertical orientation (Herrel et al., 2008). The less specialized cervicals at the dawn of modern turtle (Testudines) evolution may have permitted a transitional, intervertebral rotation of the ancestrally retracted (laterally rotated) neck towards the vertical orientation found in cryptodires by keeping the characteristic curvature of the cervical column. The full vertical swing of the curved neck inside the body wall was then enabled by the acquisition of specialized features in the anatomy of cervical eight, namely the elongated and descending posterior zygapophyses and a shortening of the vertebral centra, in cryptodires (Werneburg et al., 2015b and Williams, 1950). Transitional taxa in this regard could be the xinjiangchelyid, sinemydid, and macrobaenid stem cryptodires, in which cervical eight is not yet as specialized as in crown cryptodires (e.g., Joyce, 2007, Parham and Hutchison, 2003, Peng and Brinkman, 1993 and Zhou et al., 2014). Also, extant cryptodires show some range in the orientation of the retracted neck. It can be pure sagittal or, when enough space is available, it can be slightly rotated to the left or right body side (Werneburg et al., 2015a).

The temporal skull region of extant turtles shows a great morphological diversity. Besides the almost fully roofed coverage of the temporal region by dermal bones as found in extant marine turtles (similar to stem turtles), different modes of posterodorsal and anteroventral reduction occur among pleurodires and cryptodires alike (Jones et al., 2012, Kilias, 1957, Werneburg, 2012 and Zdansky, 1923).

Several evolutionary, functional, and anatomical factors were postulated to have influence in shaping the temporal skull region of turtles. Ten major factors were summarized by Werneburg (2012). Among those are the reduction of bones to lower skull mass, general skull dimensions, and the specific anatomy of the ear and the related jaw musculature. Among all factors, however, neck retraction has been discussed to present the most obvious correlation to skull shape (Kilias, 1957 and Zdansky, 1923). The anteroventral emargination would be associated to lateral, pleurodiran-like retraction and the posterodorsal emargination to vertical, cryptodiran-like retraction. Both emarginations would have evolved to better fit the skull into the confines of the shell during neck retraction (Kilias, 1957). Consequently, species, which lost the ability to fully withdraw the head inside the body wall (Chelonioidea, Platysternoidea), show almost no emargination and have a full bone coverage of the temporal region (Zdansky, 1923). Extinct marine turtles, such as Toxochelys (Hirayama, 1994 and Matzke, 2009) have very large emarginations. Following the consideration of Kilias (1957), I hypothesize that, in contrast to modern marine turtles, at least some stem-chelonioids were able to fully retract their necks. The same may be true for many early, extinct lineages of Testudines, which show extended emarginations [e.g., Xinjiangchelyidae, Sinemydidae, Macrobaenidae (Zhou et al., 2014), and Baenidae].

The pleurodire taxon Pelomedusidae, as a special case, shows a strong posterodorsal emargination and only a weak anteroventral one. That resembles more a typical cryptodiran than a pleurodiran skull. As such, Kilias (1957) proposed a cryptodiran-like retraction mode for pelomedusids and hypothesized that only chelid pleurodires show pure side-neck retraction. In actual fact, however, pelomedusids clearly show a typical pleurodiran retraction and not even an intermediate between the pleurodiran and cryptodiran mode (Weisgram and Splechtna, 1990 and Werneburg et al., 2015a). Zdansky (1923) highlighted that some pelomedusids have a large anterior shell opening. As such, when the head and neck are retracted, much of the skull would still be broadly exposed to predators, resulting in a covering of the “cheek” by dermal bones for further protection. All those correlations suggest that neck retraction is most important in shaping the turtle skull and that only a few obvious additional factors need to be taken into account. However, some problems remain.

The most obvious one is that almost all pleurodires and all cryptodires show both anteroventral and posterodorsal emargination – although one emargination usually dominates in the above mentioned correlation (Fig. 5). Second, the big-headed turtles (Platysternidae) do show some degree of neck retraction but the head does not pass the border of the shell to be withdrawn inside the body wall (Werneburg et al., 2015a). Nevertheless, little posterodorsal emargination is visible in the skull of this taxon. When head and carapace contact each other dorsally in such a partly retracted neck, this small emargination may have evolved to enable an optimized use of space. As mentioned above, also extant marine turtles, presumably never fully retract their necks, and also show a little posterodorsal and even a small anteroventral emargination (Jones et al., 2012). A certain, low degree of retraction, as visible during feeding in marine turtles, may result in a similar skull shape as seen in platysternids.

The results of the present study highlight that the way how the neck is retracted in cryptodires has overwhelming influence on the shape of the posterodorsal emargination in this group. Not only the shape of the curvature alone is important. Particularly the (centroid) size of the curved neck is significantly correlated to posterodorsal skull reduction in cryptodires. Usually, the neck muscles attach via tendons to the occipital skull region or they attach via broad tendon sheets to the margin of the posterodorsal emargination (Werneburg, 2011 and Werneburg, 2013b). Tendon sheets sparingly distribute the tensile forces acting on the skull. The broader the insertion of the tendon sheet (larger emargination) the better the force distribution.

A second important finding was that a certain correlation exists between the expansions of both types of emargination in all species observed. A reduction from one direction in the skull may require bone reduction of the opposite direction to retain the balanced integrity of the temporal skull region (but testing this assumption would require detailed biomechanic analyses). With the anteroventral reduction, the upper jaw and the postorbital bar lose their bony bridge to the ear capsule ventrally. However, in order to resist the tensile strains that arise during biting, the ligamentum quadratojugale is retained. It connects the upper jaw and the ear capsule ventrally (Jones et al., 2012 and Werneburg, 2013a).

The way how both emarginations are correlated with each other and the degree of bone reduction apparently are dependent on several factors such as skull dimensions (Werneburg, 2012). Some trionychid turtles with very elongated skulls (e.g., Apalone, Fig. 5C), for example, show a largely expanded posterodorsal emargination but a much smaller anteroventral emargination. The extant marine turtle skulls, with their almost fully roofed coverage of the temporal region, are more rounded in their overall shape. Several chelid pleurodires have largely expanded anteroventral emarginations and extremely flat skulls, but they lack or only show little posterodorsal emargination. Some exceptions to those intuitive correlations exist. Testing parameters of skull dimensions in relation to emargination size would be worth conducting in a different study.

Based on the detected correlation, one could argue that cryptodiran retraction mode primarily influences posterodorsal and indirectly also influences anteroventral emargination. In this regard, the low degree of retraction (i.e., slightly bended neck curve, because the head hindered performance of greater flexion) found in Platysternidae (Werneburg et al., 2015a) appears to correlate with its small posterodorsal and consequently with almost no anteroventral emargination in this taxon (Table 1, Fig. 5C). The same explanation could be true for extant marine turtles, which were not studied herein but show a similar pattern of skull construction (e.g., Jones et al., 2012).

The detected correlation to cryptodiran neck retraction nicely explains the formation of the posterodorsal emargination in cryptodires; however, no strong correlation between neck retraction and temporal skull reduction was detected for pleurodiran turtles (Table 5). An obvious interpretation would be that side-necked mode of retraction does not strongly influence head anatomy. In this regard, it is worth mentioning that the neck musculature of cryptodires drastically differs from that of pleurodires; both types are highly specialized to the unique modes of retraction (George and Shah, 1955a, George and Shah, 1955b, Herrel et al., 2008, Hoffstetter and Gasc, 1969, Jones et al., 2012, Shah, 1963, Vallois, 1920, Vallois, 1922 and Werneburg, 2011). One major difference lies in the anatomy of the musculi carapacocervicocapitis (muscular units No. 81–82 of Werneburg, 2011). M. carapacocervicocapitis lateralis Pars capitis (No. 81) originates on the first costal plate. It inserts to the skull base, namely to the prootic and opistotic in pleurodires (Shah, 1963 and Vallois, 1922) and has to be considered the major muscle for lateral retraction in pleurodires (Fig. 7D). It is completely reduced in cryptodires (except for extant marine turtles, which show only a small degree of neck retraction). The ontogenetically and phylogenetically related muscle, M. carapacocervicocapitis medialis Pars capitis (No. 82), is only present in cryptodires (except for trionychid turtles with their extremely flexible necks; Ogushi, 1913). It originates from a more anteromedian region of the carapace, particularly from the nuchal bone and inserts to the temporal fascia, which spans above the posterodorsal skull emargination (Fig. 7C; Werneburg, 2011 and Werneburg, 2013a, b). It is considered to represent one of the relevant muscles for retracting the neck in the vertical plane in cryptodires (obviously in addition to M. retrahens capiti collique Pars carapaco-basioccipitalis, No. 88, which inserts to the skull base, and other muscular units; Werneburg, 2011).

The described muscular anatomy illustrates that neck forces of lower intensity apparently act on the occiput in pleurodires than in cryptodires and that in pleurodires the stabilization of the neck is enabled by different tensile forces, which act on the skull base rather than on the occiput.

A third, non-significant but fairly large correlation to the posterodorsal emargination was found in pleurodires and cryptodires alike, namely the curvature of the ventrally flexed neck (Table 5). When the head is flexed ventrally, the dorsal neck muscles (mm. carapacocercicocapitis lateralis/medialis Partes capitis, No. 81–82; see above) and the muscles that connect the neck with the occipital skull region (i.e., musculi collosquamosum, No. 57; cervicocapitis, No. 79; collooccipitis, No. 80; see Werneburg, 2011) are stretched, which results in a tensile force on the occiput and, as shown above, fosters the expansion of the posterodorsal emargination for better force distribution.

These considerations hint to a more general view on the correlation between the unique neck and skull anatomy in turtles. The presence of the carapace in turtles and its vaulting anterior edge force the neck vertebral column to show a different orientation when compared to the dorsally convex orientation of the trunk vertebrae, which are embedded in the carapace (Fig. 7B). At resting position, either the neck has a more or less straight orientation in stem turtles or a dorsally concave orientation in crown turtles (cf. Fig. 1A; illustrated also by Werneburg et al. (2015b: Fig. S1)). Correlated with the acquisition of a shell at the dawn of turtle evolution, a reorganization of the dorsal neck musculature occurred (e.g., Gasc, 1981, Herrel et al., 2008, Lyson et al., 2013a and Werneburg, 2011). The related muscles connect the ventral face of the anterior edge of the carapace with the “sagging” neck and the skull. Different tensile forces, hence, act on the skull of turtles when compared to any possible outgroup (Fig. 7A–B). Those forces are even enlarged when the neck and head are moved during ventral neck flexion as suggested by the results of the present study. To withstand all those tensile forces – and after evolving a much longer “sagging” neck in the crown clades – the posterodorsal area of the temporal skull region must have been widened and excavated to form the posterodorsal emargination (Fig. 7B to C). This explanation could also resolve the presence of the posterodorsal emargination found in extant marine turtles, which apparently do not retract their heads inside the body.

In aquatic turtles, through the buoyancy of the water column, the gravity effect on the head and neck is eliminated. However, tensile forces on the head are still present when the turtles lay their eggs, when they bask, when hatchlings run from their nesting site to the water, and when the neck is flexed in different directions.

The tensile forces acting on the skull caused by the reorganization of the turtle trunk anatomy, i.e. the presence of a shell might be correlated to the anapsid condition of the turtle skull. Possibly, turtles plesiomorphically show an anapsid skull, which would indicate to a closer phylogenetic relationship to extinct Permian sauropsid lineages with anapsid skulls (e.g., Pareiasauria: Lee, 1997; Captorhinidae: Gaffney and Meylan, 1988; Procolophonia: Reisz and Laurin, 1991; Eunotosaurus: Lyson et al., 2010 and Lyson et al., 2013b). In that case, the emarginations were originally introduced into a primarily fully roofed temporal region.

In the case that turtles are diapsids, as indicated by several neontological and genetic studies (e.g., Field et al., 2014, Fong et al., 2012 and Tzika et al., 2011; but see Joyce, 2015), the tensile forces acting on the skull might have forced the upper temporal fenestration to be closed in the earliest stem turtles to enable a compact skull case with more evenly distributed forces on the skull than would be possible in a diapsid skull with all its thin bony bars (Fig. 7A to B; Werneburg, 2012 and Werneburg, 2013a) [side note: the same holds true for parareptilian lineages with fenestrations (MacDougall and Reisz, 2014 and Piñeiro et al., 2012) as potential close relatives of turtles]. One of the oldest stem turtles, P. quenstedti, shows a fully closed temporal region. Although the tensile forces of the neck muscles might not have yet introduced excavations into the occiput (that seems to be a derived feature of crown turtles associated to elongated neck, which results in even larger tensile forces), they may have been large enough to favor the closure of the potential upper temporal opening (Fig. 7B). The closure of the upper temporal opening and subsequently of a potential infratemporal opening (as fenestrum or cleft; e.g., compare to Evans, 2008, or Piñeiro et al., 2012) – as indicated by the herein detected correlations between the cheek and the occiput skull area – could have been facilitated by extensive broadening of the squamosal, postorbital, jugal, and quadratojugal bones (sensu Müller, 2003). Following those modifications, smaller skull elements such as the supratemporal or the postfrontal may have been reduced in order to limit the number of suture lines and hence to enable the most compact temporal bone coverage.

In this context, it is worth mentioning that neck movement, including retraction, is already detectable in the embryos of turtles (Werneburg et al., 2009). Tensile forces therefore act on the developing skull from a very early ontogenic stage and the structures of potential temporal openings might be closed already during very beginning of embryonic development. The field of experimental embryology could certainly provide more insights into this question in the future.

Given the herein detected correlation between posterodorsal and anteroventral emargination in extant turtles, the tensile neck muscle forces acting on the occiput in P. quenstedti, which like most or even all other stem turtles was terrestrial (Joyce, 2015), might have had introduced the formation of the small anteroventral emargination found in this species (Fig. 7B). The same could be true for other stem turtles with a small anteroventral emargination including Odontochelys semitestacea, the sister group of all other turtles (Li et al., 2008). Unlike all other known turtles, O. semitestacea did not have a carapace, but a plastron was developed. The condition of a half-developed shell could be ancestral for all turtles, which would mean that the plastron evolved before the carapace. However, it has been argued that O. semitestacea might have represented a bizarre early turtle lineage, which lost the carapace, possibly by paedomorphic truncation of carapace ossification, which frequently occurs in marine turtles (Reisz and Head, 2008, who actually suggested that O. semitestacea could represent the radiation of an early marine stem turtle clade), or by postmortem taphonomic processes (discussed by Rieppel, 2013, and Scheyer et al., 2013). In my opinion, specimens of O. semitestacea need a more comprehensive, monographic anatomical survey before drawing any final conclusions. In any case, the specific anatomy of the broadened ribs in O. semitestacea corresponds to certain components of a functional carapace (Lyson et al., 2014) with similar functional implications as a bony carapace (compare with the leatherback sea turtle, which secondarily evolved a functional, even more stable carapace, which, however, is anatomically not homologous to the carapace of any other turtle taxon; Delfino et al., 2013 and Rieppel, 2013). As such, comparable to P. quenstedti and other stem Testudines, tensile neck muscle forces might have favored the closure of one or two temporal openings in O. semitestacea.

Given the presented scenario and the suggested tensile forces described and if recent molecular phylogenies are correct (again, see Joyce, 2015, for comprehensive discussion of the latter issue), one may expect the future discovery of a diapsid-like stem “turtle” with an upper temporal opening and with no carapace and no (or only slightly) broadened ribs (Fig. 7A). The anteroventral area of the temporal region in this creature may not have had incipient marginal reductions as otherwise found in stem turtles such as O. semitestacea and P. quenstedti due to the lack of carapace-related tensile neck forces transferred over the occiput to the cheek region. However, given the large morphological variation of the temporal region among amniotes (synapsids, parareptiles, and diapsids; Evans, 2008 and MacDougall and Reisz, 2014), the presence of an infratemporal opening (which might anyway represent the ancestral amniote condition; Piñeiro et al., 2012), or more likely an open cleft between bones (such as in Eunotosaurus, squamates, and several early diapsids; Evans, 2008) was quite possibly present in the turtle ancestor. Finally, in this regard, also a fenestrated parareptile (see MacDougall and Reisz, 2014) as turtle ancestor could have lost its fenestration by the influence of changed dorsal muscle anatomy related to the acquisition of the shell in turtles.

The formation of the anteroventral emargination in turtles – besides the abovementioned correlation to the expansion of the posterodorsal emargination – remains unclear. At least neck movement appears to have little influence on shaping that skull region. In fact, tensile neck muscle forces, which appear to be correlated with the posterodorsal emargination, do not appear to influence the cheek region. During lateral neck flexion and during pleurodiran retraction, tensile forces certainly strongly act on the skull. However, the related muscles ventrally insert on the braincase (see above; No. 81) or insert on the ear capsule (M. plastrocapitis; Werneburg, 2011: No. 52), but not on the cheek (Fig. 7D). Certainly, some forces also act on the anteroventral area of the temporal region during sideward movements via the stretched skin (in only one turtle species a lateral muscle is known to attach to the “cheek” laterally; Werneburg, 2011: No. 44). Those transmitted forces, however, may not have such an impact on marginal reduction because they come from caudal direction whereas marginal excavation opens ventrally.

During turtle evolution, the palatoquadrate became strongly fixed to the braincase (Fig. 7B–C), resulting in the immobility of the basicranial articulation (Eßwein, 1992, Rabi et al., 2013 and Sterli and De La Fuente, 2010). One possible explanation for that pattern is that this permitted the cranium as a whole to withstand higher biting forces that might have been related to the potential herbivorous behavior in stem turtles (King, 1996 and Werneburg, 2014). For that, a mobile quadrate, as present in squamates (e.g., Iordansky, 1990), would be disadvantageous. Another, not necessarily incompatible explanation is that the palatoquadrate was fixed in order to withstand the tensile forces during neck retraction, particularly in the crown turtles with their longer neck (and stronger tensile forces) and in pleurodires in particular with their side-necked retraction mode. With the fusion of the palatoquadrate to the braincase, the temporal skull roof lost its function of stabilization of the quadrate. Consequently, bones of the temporal region could be reduced. Whereas the more posterodorsal bones may still be required as muscle attachment sites and to withstand and to buffer tensile neck forces, the anteroventral area of the temporal region may have been selected for greater flexibility, which would have favored bone reduction. In pleurodires, in which posterodorsal tensile forces are smaller than in cryptodires due to the lack of temporal skull insertion of dorsal neck muscles (Fig. 7D), the extent of this flexibility may be more pronounced in order to better fit the skull within the confines of the shell during lateral retraction (sensu Kilias, 1957). With a reduction of dorsal neck muscles in pleurodires, the anteroventral emargination can expand up to the posterodorsal part of the temporal region, whose bones finally can be fully reduced as seen in some chelids such as Chelus fimbricatus.

It has been argued that internal forces created by the jaw musculature may have a certain impact on the architecture of the temporal region (reviewed by Werneburg, 2012). Initial cladistic (Jones et al., 2012 and Werneburg, 2013b) and initial embryonic studies of the jaw musculature (Rieppel, 1990 and Tvarožková, 2006), however, did not reveal a tight correlation between jaw muscles and temporal bone arrangement. But this field apparently requires more observation.

With the fusion of the palatoquadrate to the braincase in extant turtles, which may have been initiated by the tensile neck forces (see above), the jaw musculature (i.e., m. adductor mandibulae externus; Werneburg, 2011; Nos. 17–21) is forced to bend around the ear capsule. Pleurodires bend the musculature around the processus trochlearis pterygoidei, which is more anterior than the quadrate fixation. In such way, the muscle part, which bends around that processus, can be found in a more anteriorly region of the muscle. Cryptodires, however, bend their jaw musculature around a processus trochlearis otici (Fig. 5A–B), which is formed more posteriorly by the quadrate and the prootic (Werneburg, 2013b). In both types, a part of the jaw musculature is in front and a part is behind the muscle bend. To permit maximal muscle fiber length and hence muscle force, pleurodires and cryptodires show caudally elongated supraoccipital and squamosal crests as enlarged jaw muscle origin sites. Already stem cryptodires had such crista associated to an incipiently developed processus trochlearis otici (Rabi et al., 2014). Those crests certainly increased the size of the posterodorsal emargination in various taxa. They provide a larger area for tendinous neck muscle attachments and hence, promoted as a kind of back coupling, an even more effective cryptodiran-like retraction (e.g., in trionychids). In pleurodires, a part of the external jaw musculature originates posteriorly at the ear capsule, which is an alternative or additional way of fiber elongation (Werneburg, 2013b).

The arguments presented here that explain the shaping of the temporal skull region in turtles are certainly limited due to the exclusive focus of the present study on neck curvatures. To which degree other cranial and postcranial factors contribute cannot be determined herein. It has been argued by Lyson and Joyce (2009) that some extinct baenid turtles may have evolved a full coverage of their temporal skull region in response to the predatory pressure caused by mammals after the K/Pg boundary. I will not discard this hypothesis because turtle skull shape diversity appears to correspond to several other environmental and anatomical factors than the mode of neck motion or the stabilization of the palatoquadrate alone (Werneburg, 2012).