A conspicuous feature of amniotes is the diversity of skin appendages that cover their bodies, such as hair in mammals, feathers in birds, and scales in reptiles (including the legs of birds). These appendages have been found to develop in the embryo from specialized plate-like patches of thickened epidermis, called placodes (Oliveira-Martinez et al. 2003; Milinkovitch et al. 2013; Biggs and Mikkola 2014; Di-Poï and Milinkovitch 2016). The spatial distribution of these placodes during embryogenesis is largely reflected in the distribution of the skin appendages in the body of the adult. Physical processes may also have a major role in the resulting morphological patterns (Milinkovitch et al. 2013). The resulting patterns are of great interest to zoologists. For instance, in squamates, the arrangement of scales is a character complex widely used in species delimitation and identification (e.g. Cope 1886; Campbell and Frost 1993; Weinell et al. 2019). In the herpetological literature, these scale mosaics are variously called pholidosis or scalation. When scales form large shield-like plates called scutes (from the Latin scutum, meaning “shield”) contiguous to each other, the mosaic patterns are also called scutation, or scutellation.

Here, we review different aspects of an outstanding example of evolutionarily conserved pholidosis: the mosaic of scutes of the carapace of the dorsal shell of turtles (carapace). It is a rare opportunity to be able to study the evolutionary history of patterns of epidermal appendages in deep time with abundant palaeontological data, and the turtle shell probably provides the best material of this kind.

The first known turtles with a carapace (turtles with a carapace = clade Testudinata; crown turtles = clade Testudines) are from the Late Triassic (Norian, 227-208 Ma) (Scheyer et al. 2013; Joyce 2017; Rieppel 2017). The carapace soon experienced a major change between then and the Early Jurassic (Hettangian, 201-199 Ma). After that phase, the evolutionary history of the pholidosis of the carapace (and plastron) in the majority of turtles has seen mostly minor innovations. Yet, there is substantial intraspecific variation of pholidotic patterns in turtles (Zangerl and Johnson 1957; Bujes and Verrastro 2007; Zimm et al. 2017; Szczygielski et al. 2018; Horváth et al. 2020; Maffucci et al. 2020; Bentley et al. 2021).

In this contribution we integrate a series of recent studies in palaeontology and developmental biology, including a model based on reaction-diffusion processes (Moustakas-Verho et al. 2014), to shed new light on ancient fossils and the constructional factors that have shaped the evolution of the carapace since the earliest known testudinatans.

We focus our discussion in the carapace, as the development of the plastron (ventral portion of the shell) has not been studied in similar detail. The developmental systems of scute patterning on the carapace and plastron are considered independent (Cherepanov 2014; Moustakas-Verho and Cherepanov 2015; Cherepanov et al. 2019), but with similarities (Cherepanov 2006; Moustakas-Verho et al. 2014).

Both carapace and plastron are made up of an internal layer of bone plates with contributions of the ribs, and an external layer of scutes. Secondarily, the leatherback sea turtle Dermochelys coriacea, the soft-shelled turtles (Trionychidae) and Carettochelys insculpta do not have scales on the carapace. Soft-shelled turtles are particularly divergent in their shell and its mode of development (Nagashima et al. 2012, Nagashima et al. 2014, Rice et al. 2015).

Excepting the hawksbill sea turtle Eretmochelys imbricata, there is little or no overlap between scutes. The limits between adjacent scutes form epidermal furrows termed seams, following the terminology proposed by Kordikova (2002). Each seam almost always leaves a corresponding impression on the underlying bone plates, known as a sulcus, but these may also disappear in very old adults. Thanks to the sulci it is possible to know the pholidosis of the carapace in the fossil record, despite the keratinous scutes themselves not being preserved along with the bone.

The bone plates and scute mosaic of the carapace of the vast majority of modern turtles conform to a basic plan that appeared in the clade Mesochelydia (Joyce 2017), at least as early as in the Hettangian (Early Jurassic, 201–145 Ma) (Fig. 1). This clade includes nearly all turtles known from the Jurassic to the present. The pholidotic pattern of the “mesochelydian plan” is well demonstrated for the Lower Jurassic turtle Kayentachelys aprix (Gaffney et al. 1987) (Fig. 1, Fig. 2), and can be summarized as follows: a small scute called the cervical occupies the anterio-medial border of the carapace. Posterior to it there is a series of five scutes called vertebrals, arranged along the midline. A pair of series of four scutes called pleurals are located to each side of the vertebrals. A third series is made up by the marginal scutes, 12 on each side, which are distributed along the rim of the carapace. The twelfth marginal scutes contact each other posterior to the fifth vertebral, forming together with the cervical an outer ring of scutes that surrounds the vertebrals and the pleurals. Throughout this text we will use Roman numerals to refer to the individual scutes in the mesochelydian plan, numbered from anterior to posterior. We also use Roman numerals to refer to the specific scutes of species or specimens where they can be considered prima facie homologous to the corresponding scute in the ancestral mesochelydian plan.

The underlying bone plates follow a similar arrangement that is however non-congruent with the scute pattern (Fig. 2). The antero-medial bone plate is called the nuchal, and it is followed by a longitudinal series of typically eight neural plates that develop in part from periosteal expansion of the neural arches of the corresponding trunk vertebrae (Gilbert et al. 2001; Scheyer et al. 2008). Posterior to the neurals there are typically one to three suprapygal plates, followed by the pygal plate that caps the postero-medial border of the carapace. Lateral to each neural there is a pair of costal plates, that develop from the corresponding trunk ribs by periosteal expansion and metaplastic dermal ossification (Gilbert et al. 2001). Finally, eleven pairs of peripheral plates form the lateral rims of the carapace, encircling the neurals, pleurals, and suprapygals. The tip of the ribs associated to each costal fit inside a socket in the nearest peripheral plate.

The following account synthesizes findings from Cherepanov (1992, 2006, 2016), Gilbert et al. (2001), and Moustakas-Verho et al. (2014) based on the emydids Trachemys scripta and Emys orbicularis, and the testudinid Testudo graeca. The precise timing of the events is difficult to give, because in addition to slight intraspecific and interspecific variations, authors have described the events variously in relation to incubation days, carapace length, or tabled developmental stages (Yntema 1968).

The major feature in the early development of the carapace are the “carapacial ridges”: two nearly parallel longitudinal bulges between the anterior and posterior limb buds, in the flanks of the dorsal region of the embryo (Burke 1989). Along each carapacial ridge are found the primordia of the marginal scutes, visible as placodes in the invaginations of the transverse myosepta of trunk. Similarly, the placodes of the four pleural scutes appear dorsal to each carapacial ridge, at about the level of the neural tube (Cherepanov 2006). Around the same time, in most turtles the tips of the cartilaginous ribs enter the carapacial ridge, where they become “ensnared”, deflecting their growth into the dorsal dermis (Gilbert et al. 2001; Rice et al. 2015). In trionychids, the carapacial ridge may act differently: it truncates the lateral growth of the ribs (Nagashima et al. 2012; Sánchez-Villagra et al. 2009). Both mechanisms result in preventing growth of the ribs into the ventrum of the embryo, which normally occurs in other vertebrates. The latter is a crucial step in the conformation of the overall structure of the carapace.

In subsequent stages of turtle carapace development, a series of six pairs of placodes appears along the midline, on the dorsum of the trunk of the embryo. The five posterior pairs of placodes are the primordia of the five vertebral scutes, and the anterior pair are the primordia of the cervical scute (Cherepanov 2006; Moustakas-Verho et al. 2014). Over the course of this process, the trunk of the embryo experiences significant lateral growth, especially in its middle and anterior regions, and so the elliptical outline of the carapace begins to take shape. Eventually, the primordia of all the scutes begin to expand, and as they do so, the paired primordia of the vertebral scutes merge. Concurrently, the anterior and posterior ends of the the carapacial ridges extend toward the midline, so that the left and right carapacial ridges contact each other at the posterior end of the trunk, with the primordia of the left and right marginal XII directly posterior to the primordium of vertebral V. The anterior ends of the carapacial ridges end up abutting the primordia of the cervical scute, which around this time begin to fuse. After these events, the general layout of the scutes of the carapace is in place, with an outer ring formed by the marginals and the cervical scute primordium.

The formation of the seams between the scutes has been described in detail by Cherepanov (1992), based on Emys orbicularis and Testudo graeca. The epidermis of the postero-lateral border of each vertebral and marginal scute primordia sinks sharply into the dermis, forming the first seams. Soon the same happens with the pleural primordia. The seams then proceed to expand and entirely delimit each scute.

The identities of all the carapacial scutes are thus settled, and their subsequent development is mostly concerned with changes in their proportions and further maturation of the epithelium (Alibardi and Dipietrangelo 2005). It is only around this time that the bony layer of the carapace begins to ossify, starting with the nuchal plate (Gilbert et al. 2001; Sánchez-Villagra et al. 2009), followed by the costals, and much later by the peripherals. The ossifications of the pygal and suprapygals have not been studied in detail, but begin to occur last, around the time of hatching (Emys orbicularis; Cherepanov 1992) or, more commonly, months later (Gilbert et al. 2001; Cordero 2021). The scute layer might influence the morphogenesis of the bone plates. Cherepanov (1992, 2019) noted that the seams between the marginal scutes seem to act as organization centres for the osteogenesis of the peripheral plates, by altering the local arrangement of the subjacent collagen fibres in the dermis. Indeed, in virtually all turtles (with no exceptions noted by us in the literature or in our previous experience with specimens) the intermarginal sulci fall near the middle of each peripheral plate, and this coincidence is particularly striking in hatchlings. A similar morphogenic interaction between the posterior intermarginal seam and the pygal is plausible, although to our knowledge this has not been investigated. Similar correspondences between costal plates and interpleural seams have only been reported for the Testudinidae (Procter 1922; Cherepanov 2019).

The turtle carapace provided one of the many examples that D’Arcy Thompson (1942) used in his landmark book On Growth and Form, seeking explanations of biological forms in terms of mathematical and physical processes. He classed the pholidosis of the carapace in a category of typically hexagonal tessellations that emerge as neighbouring elements squeeze each other due to divergent radial forces from within (in this case growth) or under the influence of uniform external forces. This notion does not account for the number and location of scute primordia, but remains a feasible and simple explanation for the roughly polygonal shapes, and is compatible with the more modern and detailed models that we describe below. It is likely that other authors unknown to us (e.g. Russian researchers writing in their own language) also pioneered attempts to explain the development of the pholidosis of the turtle carapace.

Over the course of the last few decades, Cherepanov and colleagues (Cherepanov 1992, 2006, 2015, 2016; Moustakas-Verho and Cherepanov 2015; Cherepanov et al. 2019) have put forward a segment-dependent model for the patterning of carapacial scutes, based primarily on histological sections of embryos and variation data. The model posits a strict correspondence between the scute placodes and thoracic myosepta. Specifically, in the standard mesochelydian plan the twelve marginal scute placodes develop in the pits of thoracic myosepta I to XII. In the most recent version of the model (Cherepanov et al. 2019), the placodes of the vertebral scutes develop at even-numbered myosepta, and the placodes of pleural scutes develop at odd-numbered myosepta, laying down the staggered pattern that is recognizable in the fully developed scutes. In the model, the association between scute placodes and myosepta also holds for deviations from the mesochelydian plan. When present, the segment-dependent model predicts that additional vertebral or pleural scute placodes appear occupying normally vacant regions of thoracic myosepta, or associated to new body segments added to the trunk. Adult scute patterns consistent with that prediction were found in the natural variation of pholidotic patterns in Lepidochelys olivacea (Cherepanov et al. 2019), and in abnormal pholidotic patterns in Testudo graeca and Emys orbicularis (Cherepanov 2014).

A causal model to explain the generation of scute patterns based on Turing patterns was introduced by Moustakas-Verho et al. (2014). A notable success in developmental biology was the introduction of mathematical models of reaction-diffusion processes by Turing (1952; Gierer and Meinhardt 1972). In their simplest form, these models describe the patterns created by the interaction of a pair of molecular agents acting as activator and inhibitor diffusing in the same medium. The activator increases the synthesis of both itself and an inhibitor, while the inhibitor suppresses the synthesis of the activator and diffuses at a higher rate. The local concentration patterns of the two molecular actors can create the so-called Turing patterns, which reproduce a variety of biological patterns, including the distribution and colouration patterns of skin appendages (Fofonjka and Milinkovitch 2021). Classical examples include the formation of spots, stripes, and labyrinths. Examples of molecular species that could act as activator-inhibitor pairs have been identified in various biological patterning systems, including the development of pholidosis in different amniotes.

The model of Moustakas-Verho et al. (2014) consists of a first reaction-diffusion process that defines the location of the scute placodes, coupled with a second reaction-diffusion process with a different pair of activator and inhibitor species. The second process describes how the primordia grow and fill the surface of the carapace. The model was implemented in a computer simulation where diffusion occurs in a rectangular grid that represents the epidermal epithelium of the trunk of the embryo. A single cell in the grid can be taken to represent a small patch of cells in the epidermis. The initial state of the model defines the chemico-kinetic properties of the activators and inhibitors, and a pre-pattern of the early location of the 12 marginal placodes along the two carapacial ridges (Fig. 3A). Additional parameters describe simple lateral growth of the thorax, its acceleration, and the timing of the onset and termination of the reaction-diffusion processes. The extension of the ends of the carapacial ridge toward the midline is not modelled.

This relatively simple model is highly successful in reproducing key features of the scute mosaic. The first reaction-diffusion process originating from the twelve pairs of marginal placodes induce the formation of two rows of four or five “pleural” placodes, and a medial row of six pairs of placodes representing the cervical and the five vertebrals. The second reaction-diffusion process mimics the growth of the primordia and the appearance of the scute seams. The result is most satisfactory in the middle region of the trunk, even reproducing the fusion of the pairs of primordia of vertebrals II, III, and IV (Fig. 3C).

The model was also successful in replicating abnormal scute patterns of actual turtles (Moustakas-Verho et al. 2014, Zimm et al. 2017). Furthermore, Moustakas-Verho et al. (2014) performed protein-soaked bead implantation experiments in embryos of Trachemys scripta that allowed them to identify Shh and FGF4 as likely molecular morphogenetic agents that may be acting as activator and inhibitor during the first reaction-diffusion process.

The reaction-diffusion model provides a causal complement to the general thrust of the segment-dependent model. This is particularly relevant because, beyond the pre-pattern of twelve marginal placodes along the carapacial ridges, the reaction-diffusion model involves no concept of body segmentation. Still, it has not been explored to what extent the reaction-diffusion model can reproduce the range of intraspecific supernumerary scale variations that are strongly suggestive of the strict correspondence between myosepta and scute placodes (Cherepanov et al. 2019). Moustakas-Verho et al. (2014) showed that axial elongation of the trunk under the reaction-diffusion reproduces a patterns of supernumerary pleurals and vertebrals, as is present in Lepidochelys olivacea. Many other patterns recorded for this species also feature supernumerary marginal scutes, which in theory represent supernumerary body segments. In any case, an explanatory bridge between the two models remains to be completed. It is possible that the transversal invaginations that mark each thoracic myoseptum participate in the reaction-diffusion process, acting as zones that facilitate the accumulation of the first activator morphogen either due to the infolding of the myosepta or due to local biochemical interactions.

There are only eight known and undisputed species of non-mesochelydian turtles, most of them from the Norian (Late Triassic, 227–208.5 Ma; except Australochelys africanus from the Hettangian) (Fig. 1), we refer to them in this text as “basal testudinatans”. The pholidosis of the carapace of only a few of those species is known in sufficient detail to remark on the differences with respect to the mesochelydian plan.

The most complex pholidosis is seen in Proganochelys quendstedtii (Fig. 2), in which one cervical, four vertebrals, four pairs of pleurals, and 16 or 17 pairs of marginals are usually recognized. In addition, Proganochelys exhibits paired series of 12 supramarginal scutes located between the pleurals and the marginals, and a so-called “supracaudal” scute occupying the caudal edge of the carapace, posterior to the fourth vertebral and flanked by the last marginals (Gaffney 1990). Many authors refer to the scute that in testudinids results from the fusion of the left and right marginal scutes XII as a “supracaudal”. To our knowledge, no author has proposed the homology between the “supracaudal” of basal testudinatans and the posteriormost marginals of testudinids.

The almost complete scutation pattern can also be observed in Proterochersis porebensis and Proterochersis robusta, where there are typically one cervical, five vertebrals, four pairs of pleurals, 14 marginals (at least 12 in Proterochersis porebensis), and three supramarginals (Szczygielski and Sulej 2016).

The left and right series of marginal scutes do not meet at the midline in the caudal region of the carapace in any of the earliest testudinatans where this condition can be ascertained: Proganochelys quendstedti, Proterochersis spp., Waluchelys cavitesta, and Palaeochersis talampayensis (Gaffney 1990; Rougier et al. 1996; Szczygielski and Sulej 2016; Sterli et al. 2021). Instead, the last marginal scutes flank a wide caudal notch (most authors call this feature a “pygal notch”, but we avoid that term because it is more commonly used to refer to a smaller notch confined to the pygal plate), and the most posterior scute is vertebral V in Proterochersis or the “supracaudal” scute in Proganochelys and Waluchelys cavitesta. This notch occurs approximately in the area of the carapace where the suprapygal and pygal plates are present in mesochelydians, and some authors have argued that therefore it is possible that basal testudinatans lacked the full complement of pygal and suprapygal plates (Sterli et al. 2021). Instead, the examination of sutural patterns in specimens of Proterochersis porebensis and Proterochersis robusta showed that the caudal region of the carapace was made up of an irregular patchwork of osteoderms wherein there are no readily apparent homologues of suprapygals, pygals, and peripherals (Szczygielski and Sulej 2019). Similar irregular sutures have been found in the carapace of Chinlechelys tenertesta (Joyce et al. 2009), of which only a few small fragments are known (Szczygielski and Sulej 2019; Lichtig and Lucas 2021).

We propose, first, that the fundamental difference in the development of the caudal region of the carapace of basal testudinatans was the failure of the lateral marginal scute series to meet at the midline, posterior to the primordium of vertebral V. It is easy to derive this inference from the fact that in extant turtles the series of marginal scute primordia develop early along the carapacial ridges, when the ridges are roughly parallel to each other, on the flanks of the embryo. Thus, vertebral V remains at the postero-medial edge of the carapace, preserving the relative position of the scute primordia from the early stages of development. This leads us directly to a second proposition: that it might not be necessary to retain the notion of a distinct kind of scute called “supracaudal”, unique to basal testudinatans. Instead, we propose that the “supracaudal” attributed to Proganochelys and Waluchelys might be a very short vertebral V, with the same topological relations as seen in Proterochersis, except with respect to the supramarginals. Embryological observations and the reaction-diffusion model show that the all scutes along the midline of the carapace (the cervical and the vertebrals) have the same initial mode of development from paired placodes (Cherepanov 1992, 2006; Moustakas-Verho et al. 2014, Moustakas-Verho and Cherepanov 2015). The “supracaudal” can also be expected to belong to the same meristic series, based on its position in posterior end of the midline of the carapace. That position in the series of midline scutes in Proganochelys indicates identity with vertebral V. The morphology of the “supracaudal” is clearly different from vertebrals I–IV, but it does resemble the morphology of the cervical, which is another specialized scute in the same meristic series. Also like the cervical, and unlike the other vertebrals, the “supracaudal” scute in Proganochelys (and apparently Waluchelys) does not contact pleural scutes. This condition is, to our knowledge, unprecedented in other turtles, but it does emerge in simulations with the reaction-diffusion model (Fig. 3C; see more details below).

Thus, in our new interpretation, Proganochelys has five vertebral scutes, just as Proterochersis and the vast majority of mesochelydians. The pholidosis of the central region of the carapace of other basal testudinatans is unknown, and in the absence of contradictory evidence, it is reasonable to presume that they also had five vertebral scutes in total.

Another possible interpretation of the nature of the “supracaudal” of basal testudinatans, is that it is the result of the fusion of multiple marginal scutes, and therefore the marginal series meet in the caudal region of the carapace in Proganochelys and Waluchelys. A fusion of the XIIth pair of scutes occurs in tortoises (Testudinidae), which results in a single scute occupying the posteromedial edge of the carapace (see above) (Joyce and Bell 2004; Cherepanov 2006). However, testudinids do not feature comparable caudal notches, and this alternative interpretation is less parsimonious with respect to the relations of homology. Moreover, there are analogues in other clades that suggest the feasibility of the anatomical configuration that we propose. Among extinct mesochelydians, the cryptodirans Sinemys and Eurysternum wagleri, and various species in the clade Baenodda (Fig. 4) also display caudal notches (Tong and Brinkman 2013; Anquetin and Joyce 2014; Joyce and Lyson 2015), with vertebral V occupying the posterio-medial edge of the carapace and flanked by the left and right series of marginal scutes that do not meet. We posit that those species represent very rare instances of re-evolution of the basal testudinatan caudal region, with a similar developmental origin.

Apart from the caudal notch, another distinctive feature of basal testudinatans is the presence of supramarginal scutes. This is the most difficult series of scutes to characterize, as their numbers and anatomical location vary widely across species. In addition to basal testudinatans, supramarginal scutes also occur in the alligator snapping turtle Macrochelys temminckii (Pritchard 1979; Joyce 2016), and in fossil eubaenines (Joyce and Lyson 2015) and the stem-pleurodire Platychelys oberndorferi (Sullivan and Joyce 2017) (Fig. 4). The reaction-diffusion model can recreate additional longitudinal series of scutes by altering the diffusion area or the diffusion rates of the activator and the inhibitor in the first reaction-diffusion process (Zimm 2019). We explored adjustments with both approaches, and obtained the most satisfactory results by halving the diffusion rates (Fig. 3B), while keeping all the other model parameters identical to the values used by Moustakas-Verho et al. (2014). The control files for reproducing the simulations are provided in the supplementary files. The result roughly approximates the pholidosis of Proganochelys (Fig. 3C), but also introduces several supernumerary scutes in the vertebral and pleural series. It might be possible to obtain more realistic results with other parameter combinations, or with further adjustments of the model. However, the location and number of supramarginal scutes is highly variable between species (Fig. 4, Table 1), and likely involve non-homogenous developmental conditions not yet considered. The ideal path for researching supramarginal scutes would be to study of the early development of the carapace in Macrochelys temminckii, the only extant species that features supramarginal scutes.

The evolution of the testudinatan shell from the Triassic to the present has resulted in a net reduction in the average number of its constituent elements. Various authors have thus recognized this as a macroevolutionary trend towards simplification of both the numbers of bony plates and epidermal scutes (Gadow 1899; Newman 1906; Zangerl 1969; Gaffney 1990; Cherepanov 2015; Cordero and Vlachos 2021). Zangerl (1969), for instance, conceived progressively simplified “levels of organization” of the testudinatan shell, which different lineages of turtles would have traversed in parallel.

Here we take a closer look at the history of gains and losses of carapacial scutes. We will first consider the majority of turtles that retain the epidermal scutes. We will address turtles with complete loss of scutes separately.

The modern turtle shell displays a distinctive pholidotic pattern in that a large surface of the body is covered with a mosaic of few scales (typically 38 in the carapace; Fig. 2). We compiled a list of independently accumulated gains and losses of carapacial scutes from basal testudinatans to the present (Table 1). The list is not exhaustive. We focused on documenting gains and losses that were unambiguous given the current incomplete knowledge of the phylogeny of testudinatans, and in including the most extreme cases. The relationships between basal testudinatans are unresolved (Sterli et al. 2021), so we did not count gains or losses among them, and we considered both Proganochelys quendstedti and Proterochersis robusta as representing possible ancestral states for counting the losses accumulated along the lineage leading to the Mesochelydia. The list shows that variations with fewer scutes are more widespread among extant species (Testudinidae 65 spp., Pelomedusoides 35 spp., and Kinosternidae 27 spp.; Turtle Taxonomy Working Group 2017). However, when the independently accumulated scute gains and losses are tallied, the gained scutes are more numerous (Table 1). This apparent contradiction results from the extinction of many clades with scute gains, and the great diversification of a few clades with scute losses. But is this historical pattern the result of chance, or are there distinct processes consistently driving this dynamic over macroevolutionary time scales? If so, what are they? Even though the direction and trajectory of the mean number of carapace scutes appears downward in time, the presence of a trend in the sense of a recognizable statistical bias has not been tested. A rigorous approach to these questions ought to make use of comparative phylogenetic methods, which provide means to study hypotheses about macroevolutionary regimes with explicit stochastic models (Ascarrunz et al. 2019). Dynamics that produce within-lineage scute-gaining trends can be characterized with Brownian motion with a bias (called Brownian motion with a trend; Hunt 2006), or Ornstein-Uhlenbeck processes (Harmon 2018). A between-lineage trend toward scute loss can be represented by the SSE (State Speciation and Extinction; Maddison et al. 2007) family of models where the value of a trait (number of scutes) are associated with differential rates of speciation and extinction (Harvey and Rabosky 2018; Herrera-Alsina et al. 2019; Nakov et al. 2019).

Cordero & Vlachos (2021) presented the first quantitative analyses of the evolution of the numbers of shell elements with comparative methods. But they too acknowledge that scute gains and losses represent fairly rare events over the course of the evolutionary history of testudinatans, and that it is difficult to sample the relevant species in an unbiased manner. For this reason, the overall rate of change in scute number is bound to be small. If there are statistically identifiable trends in the dominant macroevolutionary regime, they are subtle. Properly characterizing them will require extensive scute number variation data, and a comprehensive time-scaled phylogeny of testudinatans. Neither are available at the moment.

Finally, we note that it is difficult to conceive a distinct and plausible mechanism for between-lineage downward trends in scute number. Between-lineage dynamics have been attributed to species selection (Jablonski 2008), where the value of a trait affects the rates of speciation and extinction specific to each lineage. Rather than diversification being affected by gains and losses of scutes, it seems more plausible to assume the presence of unknown developmental factors that increase the liability to scute gain but also have other morphological or physiological effects that decrease fitness or adaptability the population. Cordero & Vlachos (2021) are also hesitant to associate trends of loss of bone and scute elements of the shell in general with specific mechanisms. Given the paucity of data, these matters are likely to remain in the realm of speculation.

It has been suggested that groups of species in a new “Bauplan” have more variability (i.e., more capacity to generate variation) than subsequent clades, which are less prone to vary (Waddington 1957). This pattern has been associated with “canalization”, defined as “the reduced sensitivity of a phenotype to changes or perturbations in the underlying genetic and nongenetic factors (e.g., the environment) that determine its expression” (Flatt 2005: 288). It is thus hypothesized that morphological variation among species appears before taxonomic diversity (Erwin 2007). As in most other vertebrate clades examined using palaeontological data (Benton et al. 2014), in turtles there seems to be more variation and “morphological experimentation” in the early evolution of the group than in more nested clades as the scute pattern is concerned. An issue that remains to be examined in basal testudinatans is that of intraspecific morphological variation in macroevolutionary timescales, and whether it reflects canalization (Pimiento et al. 2018). The limited number of known species and lack of sufficient sampled individuals fail to provide a proper dataset for a quantitative assessment of the matter.

The evolutionary history of testudinatan carapacial scutes involved different macroevolutionary regimes. We examine in the following subsections the clades that feature complete loss of shell scutes: Trionychia, and the marine turtles Dermochelyidae and Protostegidae. The developmental processes that produce total loss of scutes (see below) are quite different from the ones that produce the loss of individual scutes, where the total corneous coverage of the shell is preserved by fusions or compensatory expansions of other scutes in a space-filling manner (reviewed in Cherepanov 2015). Also, the resulting external shell skin is unlike the cornified skin of the shell of other turtles. Thus, they are likely linked to developmental and adaptive shifts that delineate distinct macroevolutionary regimes.

The clade Trionychia includes the Trionychidae (softshell turtles) and Pan-Carettochelys (the scute-less pig-nosed turtle Carettochelys insculpta and numerous extinct relatives; Joyce 2014). These are highly aquatic turtles, in which the importance of water-retaining capacity of cornified scutes (Alibardi 2003) is reduced, and there is a physiological advantage in the utilization of the unkeratinized skin on the surface of the shell as a respiratory organ (Scheyer et al. 2007). Trionychids have evolved towards efficient and prolonged extrapulmonary aquatic respiration (Ultsch et al. 1984; Stone et al. 1992), an important part of which can occur through the skin (ca. 70% for Trionyx triunguis in an experimental setting; Girgis 1961). In particular, the skin covering the plastron in the trionychid Apalone mutica is much more permeable than the shell scutes and other skin in other aquatic turtles (Dunson 1986). Cutaneous respiration in Carettochelys has not been studied directly, but its shell skin is grossly similar to that of trionychids. Carettochelys inhabits predominantly normoxic waters, which is consistent with important skin function in gas exchange (Joyce et al. 2012). There are also quantitative similarities in porosity of costal plates of Carettochelys and trionychids, which is a proxy for the vascularization associated with cutaneous respiration, but this relationship is not unequivocal (Clarac et al. 2020). Finally, there is significant evidence for parallel adaptation in genes critical to aerobic respiration (oxidative phosphorylation pathway genes) in the mitochondrial genomes of trionychids and Carettochelys (Escalona et al. 2017), underscoring the evolutionary importance of physiological modifications of the respiratory systems in the two lineages.

We infer that in early trionychids and the ancestors of Carettochelys, scute loss was driven by the adaptive value in the decornification of the skin covering the shell, and a relaxation of the strength of natural selection for the maintenance the developmental system of periodic patterning that determines the scute mosaics. Similar hypotheses have been put forward by previous authors (Scheyer et al. 2007; Nakajima et al. 2017). The fossil record of Pan-Carettochelys (Joyce 2014) and trionychids (Georgalis and Joyce 2017) clearly shows that scute losses occurred in parallel, after the divergence of the two clades from their last common ancestor. The ultimate mechanisms by which scute loss is realized in both lineages are also quite different. Juveniles of Carettochelys insculpta display rudimentary and apparently unkeratinized scutes on the carapace, with visible seams and an unusual pholidosis of as many as 8 vertebrals, a cervical, and marginals (Zangerl 1959; partially visible in the photographs in Beggs et al. 2000). The scute seams disappear completely by adulthood. Thus, the basic periodic patterning system is retained in Carettochelys, although the induced seam pattern has been allowed to drift away significantly from the mesochelydian plan. The transient occurrence of intermarginal seams might be necessary for the development of the peripheral plates, per Cherepanov’s (2019) observation that these seams act as organizers of peripheral plate ossification centres (see above). Trionychids, in contrast, have dispensed with both seams and peripherals. Among trionychids, peripheral-like ossicles develop only in Lissemys spp., and their homologies are still uncertain (Delfino et al. 2010); all other species completely lack peripheral-like structures. In the trionychid Pelodiscus sinensis there is no formation of the initial patterning of marginal scute placodes and the correlated punctuated expression of Shh, Bmp2, and Gremlin along the carapacial ridges, preventing the subsequent induction of the rest of the scute mosaic (Moustakas-Verho et al. 2014).

It is more difficult to hypothesize about the drivers for scute loss in the marine turtle clades Protostegidae and Dermochelyidae. Only the latter is represented by an extant species, the leatherback turtle Dermochelys coriacea. The Protostegidae might be stem-chelonioids or the sister clade of Dermochelyidae (Evers and Benson 2019; Evers et al. 2019a); in either case there is palaeontological evidence showing that the loss of scutes occurred independently in both clades (Zangerl 1960; Hirayama and Chitoku 1996; Hirayama 1997). Cutaneous respiration is not a feasible strategy for marine turtles, and for Dermochelys coriacea in particular (Escalona et al. 2017). The loss of cornified scutes and subsequent modifications of the shell in dermochelyids might be associated with the origins of many adaptations to life in open sea and deep diving, as seen in living leatherbacks. The lack of a fully rigid shell, particularly in the plastron, probably assists leatherbacks in accomplishing lung collapse during diving (Fossette et al. 2010). The presence of a thick subcutaneous layer of blubber probably also assists deep diving (> 1000 m) by providing thermal insulation, and can serve as a nutrient reserve during long migrations. The elasticity of the uncornified skin allows significant alteration of the thickness of the blubber deposits and overall body shape in response to ecophysiological factors, such as nutrient availability and reproductive status (Davenport et al. 2011). Although those facts alone do not amount to a strong case for diving and thermoregulation as the specific drivers of scute loss, they show that Dermochelys displays extensive adaptations to a unique physiology and habitus among living turtles, which may justify the presumption of a distinct evolutionary regime is operating on dermochelyids. The protostegid Desmatochelys has enlarged semicircular canals, a feature that is even more pronounced in Dermochelys coriacea (Evers et al. 2019b), which could indicate similarities in locomotor behaviour between the two species, although the precise functional implications remain unclear (S. Evers, pers. comm. 2021).

It is illuminating to seek parallels between character systems in different clades. For the pholidosis of the turtle carapace, though, is difficult to come by with analogous (or homologous) character systems that display relevant similarities in the morphological, developmental, and evolutionary characteristics that we have discussed in this paper. For instance, the large scales on the body of pangolins do not grow in a surface-filling fashion forming a mosaic, as they are arranged in numerous overlapping rows (Wang et al. 2016). Armadillo armour scales are more similar to turtles’ in that they do not overlap and are closely associated to an underlying surface of osseous plates, but these scales are minute (Chen et al. 2011), and it is unclear if individualized elements can be recognized across specimens or species.

We identified a more satisfactory analogue in a different anatomical region. There is great diversity in the pholidotic patterns of the head in squamates, including mosaics of scales of a wide range of sizes, from proper scutes to small “granular” scales. In squamates excluding gekkotans and dibamids, scutes are fairly common in the “pileus region”: the dorsal surface of the head extending from the tip of the rostrum to the occiput. Notably, when scutes are found in the pileus, they have a tendency to be form similar arrangements, and individual scutes can be recognized between species (Ursel 1978; Estes et al. 1988) (Fig. 5).

In a manner roughly analogous to the mesochelydian plan, a stereotypical pileus scute pattern has been considered ancestral for squamates (Cope 1898; Ursel 1978; Estes et al. 1988), but different clades may have slight variations of the ancestral pattern, and the homologies and evolution of the pileus scutes across squamate clades have not been comprehensively examined in a modern phylogenetic framework (Gauthier et al. 2008). Similar patterns of homologous cranial scutes are found in turtles, but this character complex remains even more understudied in the majority of turtle clades. Head scutation is used for the identification of marine turtles (Wyneken 2001), and has also been described for a limited number of stem turtles (Simpson 1938; Sterli and de la Fuente 2013; Evers et al. 2021).

Based on photographic data collected for a previous study (Esquerré and Keogh 2016), we hypothesize the loss and re-evolution of a typical pileus scute pattern within snakes. Boas and pythons (Boidae and Pythonidae, or Booidea and Pythonoidea of some authors), are two closely related clades, the Constrictores (Reynolds et al. 2014; Burbrink et al. 2020; Georgalis and Smith 2020) Species in these clades display pileus scale patterns ranging the scutes that are typical in most snakes to granular scales with little differentiation. The latter condition is found in the majority of boa species. Early studies alternatively suggested that granular scales in the pileus represent the plesiomorphic (Zacharias 1898) or the derived (Werner 1899) condition for these clades. The modern phylogenetic understanding of snake phylogeny (Reynolds et al. 2014; Burbrink et al. 2020) suggests that medium-sized scales or large scutes in the typical snake pattern are likely primitive for both boas and pythons, but the typical scute pattern is also found deeply nested within boas, indicating its secondary acquisition (Fig. 5).

The head is a complex anatomical structure that tightly accommodates and supports diverse sense organs and mechanical feeding specializations, among other functions. Turtle shells, in contrast, provide a vessel for a large body cavity where organs are arranged more freely, and are routinely displaced due to head and limb tucking, food intake, and gravidity. Unlike bones, most specific scales are not intimately associated with other critical components of the head (except the interoccipital scale with the pineal eye in lizards, and scales surrounding eyes and nares), and therefore the topography of the scutes should be more prone to vary with respect to the geometric and mechanical diversity of the vertebrate head. A worthy avenue of research would be to examine if the evolution of pholidotic patterns and scale size reflect the parallel ecological adaptations seen in the gross head geometry of pythons and boas (Esquerré and Keogh 2016). Ursel (1978) hypothesized that small scale size in the pileus facilitates increased cranial kinesis.

We infer that the scale systems of the turtle carapace and the pileus of squamates are subject to similar kinds of developmental biases (Uller et al. 2018), that strongly favour a narrow range of pholidotic arrangements when the scales are large and few. These biases are likely related to the intrinsic properties of the postulated reaction-diffusion processes that are involved in the patterning of skin appendages of the turtle carapace (Moustakas-Verho et al. 2014) and the body of amniotes in general, with some evidence of also being in place in the squamate head in particular (Milinkovitch et al. 2013), and indeed in the repeated skin appendages of vertebrates in general (Dhouailly et al. 2019). If this is correct, we postulate a highly conserved “pre-pattern” of scale placodes on the head of squamate embryos, and a highly conserved general system of periodic patterning. The latter is expected if evolution has difficulty traversing between distant parameter regions that yield viable Turing patterns (Scholes et al. 2019).

In turtles, a release from the bias towards the mesochelydian plan occurs in trionychians and Dermochelys, where the lack of scutes obviates the need for preserving a patterning morphogenetic process (Moustakas-Verho et al. 2014) (see above). However, carapacial scutes never significantly increase in number and reduce in size as the head scales of squamates do. It can be assumed that the pholidosis of the turtle carapace derives from a more traditional reptile pattern with numerous small scaled covering the trunk of the body, and for some unknown mechanism caused the canalization of the patterns of basal testudinatans (canalization occurred again, to a lesser degree, with the origin of the Mesochelydia). We found in simulations with the reaction-diffusion model that by reducing diffusion rates, we can induce far greater numbers of scutes than in existent testudinatans, with a limit (Fig. 3B). Of course, in these simple experiments with an idealized model we are free to adjust parameters in ways that are not necessarily accessible to evolution due to physical limits and interactions with other developmental systems.

Unfortunately, the fossil record is unlikely to shed light in the matter. Unlike the bony shell (Li et al. 2008, 2018; Lyson et al. 2013; Schoch and Sues 2015), the early stages of the evolution of the shell scutes have not been documented in the fossil record in non-testudinatan stem turtles. For the sulci to form it is necessary a very close interaction between the epidermis and the bone surface, which likely did not occur in the earlier ancestors of turtles.

Developmental models aid palaeontologists to assess problems about character change in explicit causal frameworks. Even if the models do not suffice to explain all the relevant variation, they can shed light on matters previously only understood as far as the traditional approach of pattern-matching can reveal. In the carapace, the recent discoveries highlight the relationship between body segmentation and scute patterning, and how the same structures in the flanks of the carapace, –the carapacial ridges–, contribute to the formation of the outer ring of the carapace while possibly inducing its internal integumental patterning, as suggested by the temporal and causal priority of the marginals over the pleurals and vertebrals. We showed how the origin of the mesochelydian plan can be understood in these terms.

For the vast majority of turtles that retain their scutes, the macroevolutionary patterns are more complex and subtle than what might be conveyed by the “trend toward scute loss” that is traditionally suggested in the literature. Still, the reasons why scute number evolution remains highly constrained remain unknown, especially in contrast with what it is possible to induce in the developmental models, and also with what is seen in the diversity resulting from a similar developmental system in squamates.

On a more general note, we hope that this paper is an example of how considerations of ontogeny offer a deeper and necessary understanding of morphological transformations that occur in macroevolutionary time (Maier 2021). Our understanding of turtle evolution has benefited by work inspired by Wolfgang Maier, as in fundamental considerations of cranial anatomy (Werneburg and Maier 2019) or the evolution of the pleurodire shell (Scheyer et al. 2008).