The simplification of the tetrapod skull occurred convergently in various tetrapod lineages from the Devonian into the Mesozoic, leading to some groups (e.g. lissamphibians) to retain only 19 of the original 41 dermal roof bones present in stem-tetrapods. Despite the potential to shed light on the functional adaptations and developmental mechanisms behind skull simplification, little work has been done on the distribution of bone loss across tetrapod phylogeny. We conducted maximum likelihood ancestral state reconstructions for the presence/absence of temporal and median dermatocranial bones using two large composite trees that placed Lissamphibia either within Temnospondyli or Lepospondyli, reflecting the ongoing debate on lissamphibian origins. Our results indicate that the temporal series did not form a developmental module, as the loss of these bones was quite variable. With the exception of Sauropsida, the intertemporal bone was lost first, followed by the supratemporal, and then the tabular and/or postparietal. In Sauropsida, the tabular and/or postparietal was the second bone to be lost. The supratemporal was lost and regained repeatedly, and was found to be the most variable element, while the nasal, frontal, and parietal were the least variable. Interestingly, the ontogenetic timing of ossification does not correlate with the propensity for a certain bone to be re-acquired or lost. No obvious relationship was found between skull simplification and lifestyle or body size. In summary, the simplification of the dermatocranium is a more complex process than previously thought, and likely involved a mixture of developmental, ecological, and functional drivers.

Ancestral state reconstruction, dermatocranium, lepospondyl, lissamphibians skull simplification, temnospondyl, tetrapod

The skull is one of the most complex and versatile components of the vertebrate body plan. In addition to housing the sensory organs that allow vertebrates to navigate and interact with the world, this anatomical unit is also used in feeding (Heiss et al. 2018), locomotion (Wake 1993), combat (Farke et al. 2009), and sexual display (Knell et al. 2013). The main components of the skull include the (1) endocranium (chondrocranium and neurocranium) which protects the brain, the (2) viscerocranium (gill arches and splanchnocranium) which originates from the branchial arches, and (3) the dermatocranium, which acts as a vault for these structures (Walker et al. 2000). The complexity and multifunctional nature of the skull is a direct result of selection pressures applied to these individual components over evolutionary time.

One of the most significant moments in the evolution of the skull was the water-to-land transition in Tetrapodomorpha (the clade made up of extant tetrapods and extinct species more closely related to them than to lungfish) during the Devonian, which was a time of rapid change, morphological innovation, and ecological radiation (Long and Gordon 2004; Clack 2009, 2012). Previous studies have already outlined some of the morphofunctional shifts in the skull during this period, including the evolution of the tetrapod otic region, and adaptations associated with feeding and breathing on land (Coates and Clack 1991; Clack 1992, 1994, 1998; Clack et al. 2003; Brazeau and Ahlberg 2006; Schoch and Witzmann 2011). Another key change that occurred at this transition was the apparent reduction of bone elements in the dermal skull. This can first be seen in Acanthostega and Ichthyostega which lack the extrascapular and operculogular bones present in finned tetrapodomorphs such as Eusthenopteron (Andrews and Westoll 1970; Jarvik 1980). The loss of the extrascapulars was a pivotal moment in tetrapodomorph evolutionary history, marking the origin of the neck as the skull became detached from the dermal pectoral girdle (Clack 2002; Sefton et al. 2016; Maddin et al. 2020) (note, a functional ‘neck’ has been reported in the tetrapodomorph fish Mandageria fairfaxi (Johanson et al. 2003)).

The trend towards a simplified skull continued after the Devonian and can be observed throughout tetrapod evolutionary history, occurring independently (i.e. convergently) in several lineages with a diversity of skull morphologies and functional adaptations (Rawson et al. 2022) (Fig. 1). Skull simplification has been extensively documented in reptiles (Williston 1925; Lee et al. 2020), birds (Smith-Paredes et al. 2018; Plateau and Forth 2020; Asakura and Kawabe 2022), synapsids (Sidor 2001), and lissamphibians (caecilians, salamanders, and frogs), with the latter only retaining 19 of the original 41 bony elements found in many stem-tetrapod skulls (Fig. 1; Shoch 2014). Interestingly, the convergent bone loss across tetrapod lineages is not only seen in the overall number of elements lost, but also in the consistent loss of particular bones, the intertemporals being a prime example (at least as separately ossified elements). In contrast, bones such as the parietals have been conserved for over 400 million years from the first osteichthyans to extant tetrapods (Fig. 1). This would suggest that there is a commonality in the evolutionary pathways towards skull simplification, regardless of morphofunctional differences in cranial construction. Surprisingly, the distribution of bone loss across tetrapod phylogeny has received little attention. Research has instead been largely focused on quantifying simplification in terms of changes in complexity through network analyses (Esteve-Altava et al. 2013, 2014; Lee et al. 2020; Plateau and Forth 2020; Asakura and Kawabe 2022; Rawson et al. 2022; Strong et al. 2022), and determining the mode of loss in extant taxa through fate mapping (Maddin et al. 2016), histological (Arnaout et al. 2022), and embryological studies (Koyabu et al. 2012; Smith-Paredes et al. 2018).

Figure 1. 

Summary of the changes in composition of the median and temporal series in tetrapods. For a more detailed overview of the presence/absence of median and temporal series elements see Fig. 2. (Stem tetrapod = Eusthenopteron (Carroll 1988); Temnospondyli = Apateon (Schoch and Milner 2014); Salientia = Bombina (Maglia and Púgener 1998); Caudata = Cryptobranchus (Elwood and Cundall 1994); Gymnophiona = Ichthyophis (Wilkinson et al. 2014); Seymouriamorpha = Seymouria (Berman et al. 2000); Diadectomorpha = Tseajaia (Berman et al. 1992); Synapsida = Eothyris (Reisz et al. 2009); Therapsida = Suminia (Rybczynski 2000); Captorhinidae = Captorhinus (Fox and Bowman 1966); Protorothyrididae = Paleothyris (Carroll 1969); Araeoscelidia = Petrolacosaurus (Reisz 1981); Varanopidae = Mesenosaurus (Reisz and Berman 2001); Basal Neodiapsida = Weigeltisaurus (Pritchard et al. 2021); Archosauriformes = Alligator (Jollie 1962); Rhynchocephalia = Sphenodon (Jones et al. 2009); Squamata = Cordylus (Evans 2008); Parareptilia = Macroleter (Tsuji 2006); Recumbirostra = Eryodus (Carroll and Gaskill 1978); Nectridea = Diplocaulus (Carroll et al. 1998); Aïstopoda = Pseudophlegethontia (Anderson 2003).

At the developmental level, bone loss in the skull is known to occur in one of two ways: either through (1) fusion or (2) lack of ossification. Fusion of adjacent bones can occur when ossification centres merge with neighbouring ones early in development due to the premature closure of suture joints, or when the cartilaginous anlagen fuse before the suture joint develops. This then forms a composite bone which is made up of multiple ossification centres, and thus gives the impression that a bone element has been lost (Koyabu et al. 2012; Esteve-Altava et al. 2013; Schoch et al. 2014). The mammalian interparietal is an example of such a composite bone, consisting of four fused ossification centres thought to be homologous to the postparietals and tabulars (Koyabu et al. 2012). The second developmental mechanism of skull bone loss is the failure of ossification centres to form, often caused by heterochronic truncation. This typically affects those bones that form in the final stages of ancestral skull ossification, as the shortened ontogenetic trajectory prevents them from forming (Schoch 2014). In other words, ossification fails to occur in bones which, primitively, tend to ossify in the later stages of skull development. The missing dermal bones in the skulls of extant lissamphibians, for instance (e.g. the postfrontal, postorbital, and tabular), are those that form late in the ontogeny of temnospondyls (Schoch 2002), the putative stem group of Lissamphibia (Carroll 2007; Ruta and Coates 2007; Anderson 2007, 2008; Schoch et al. 2020). However, dermal bones can also fail to ossify if the minimum population size of stem cells in the mesenchymal tissue layer where these bones form is not reached (Atchley and Hall 1991), as may be the case in instances of miniaturised body size (Pérez-Ben et al. 2018; see below).

Various functional adaptations have been proposed as potential evolutionary drivers of cranial simplification. For example, it has been hypothesised that a reduction in the number of bony elements would increase the degree of connectivity in the individual remaining cranial bones (Strong et al. 2022), which in turn would lead to a more robust skull capable of withstanding the biomechanical stresses associated with certain lifestyles such as crevice dwelling (e.g. in extant lepidosaurians) (Herrel et al. 2007) and head-first burrowing (e.g. in recumbirostran microsaurs, caecilians, amphisbaenids, dibamids, and scolecophidian snakes) (Pardo et al. 2015; Szostakiwskyj et al. 2015; Pardo and Anderson 2016; Strong et al. 2022). In the case of batrachians (frogs and salamanders), Schoch (2014) suggested that the skull morphology and bone reduction may have been driven by the rearrangement of jaw musculature related to skull flattening. The study proposed that a highly fenestrated skull morphology and the emargination of the cheek region in salamanders and frogs (via the loss of the postfrontal, postorbital, jugal, postparietal, supratemporal, and tabular) would allow for an extension of the muscles onto the dorsal surface of their flattened skull. In synapsids, it has been hypothesised that skull simplification occurred as an adaptation to strengthen the skull in response to biomechanical forces applied to the skull roof by jaw abductor muscles (Sidor 2001). An alternative argument was presented by Koyabu (2023) who posited that skull simplification in synapsids may be the by-product of increasing brain size over the evolution of the clade instead of a functional adaptation. Evidence supporting this hypothesis comes from the presence of genes such as Dlx5 and Lmx1b which are known to influence both brain expansion and suture closure in the skulls of mammals (Koyabu 2023).

Miniaturisation is another developmental change linked to skull bone loss; this is the evolution of an adult body size so small that it surpasses a threshold at which dramatic changes in morphology, physiology, and ecology can occur (Hanken and Wake 1993; Yeh 2002; Pérez-Ben et al. 2018). Different explanations have been given to link extremely small body sizes to skull simplification. On the one hand, bone loss might occur as a by-product of ontogenetic truncation, which seems to be a frequent mechanism for the evolutionary reduction of body size, as miniaturised species typically resemble the juvenile stages of related non-miniaturised taxa (Hanken and Wake 1993). In this regard, miniaturised species lack bones that form late in the ontogeny of these larger species (Hanken and Wake 1993; Yeh 2002; Schoch and Rubidge 2005; Schoch 2013a; Scherz et al. 2019). Whereas skull simplification by ontogenetic truncation is related to the causes of miniaturisation, the small body size itself may impose developmental constraints that lead to bone loss (Yeh 2002; Pérez-Ben et al. 2018). As discussed above, it has been suggested that the minimum number of mesenchymal cells required to induce bone formation cannot be reached at extremely small body sizes, thus preventing bones from ossifying (Atchley and Hall 1991; Pérez-Ben et al. 2018).

Physical and functional constraints may also play a role. For example, certain structures, such as the eye, require a minimum size to be functional and are therefore proportionally much larger in small species. Consequently, the need to retain functionality in a sense organ might lead to major morphological rearrangements of the skull to accommodate the organ within a miniaturised structure. One instance of this is found in geckos, which are thought to have lost their postorbital and supratemporal because of the space taken up by their large eyes (Herrel et al. 2007). Together, all these proposed evolutionary drivers highlight the complexities of skull bone reduction and show that the evolutionary underpinnings of this trend might not be uniform among clades.

In addition to providing insight on the functional and developmental drivers of skull evolution, the simplification of the dermal skull roof has also been used in phylogenetic studies that address the relationships of early tetrapods. Many such studies have utilised the absence of specific cranial roof bones as ‘loss characters’ which have been central to ongoing discussions surrounding the origin of lissamphibians. Currently there are two main lines of thought: (1) the Temnospondyl Hypothesis (TH) which is favoured by most authors and states that lissamphibians form a monophyletic group within temnospondyls and are nested within Dissorophoidea, likely with Amphibamidae (Anderson 2007, 2008; Carroll 2007; Ruta and Coates 2007; Schoch et al. 2020) but Branchiosauridae have also been suggested as putative lissamphibian relatives (Milner 1993; Anderson 2007; Carroll 2007; Ruta and Coates 2007), and (2) the Lepospondyl Hypothesis (LH) which suggests that a monophyly is formed between lissamphibians and lepospondyls (Laurin 1998; Vallin and Laurin 2004; Marjanović and Laurin 2008, 2009; Laurin et al. 2022). There are some variants of the TH, Pardo et al. (2017a), for instance, hypothesised that caecilians are derived from stereospondyls, while frogs and salamanders are nested within dissorophoids. Previously, a third hypothesis had also been considered to explain lissamphibian origins, namely the Polyphyly Hypothesis, which placed frogs and salamanders with temnospondyls, and caecilians with lepospondyls (Carroll 2001, 2009; Anderson et al. 2008a; Lee and Anderson 2006); this has received very little attention, though, in more recent analyses.

The first step to unravelling the evolutionary factors that underlie the convergent evolution of bone loss is to identify any patterns of loss and regain. This presents a challenge, however, as determining whether or not a bone has been truly lost or has simply fused with others is often impossible, in particular when dealing with fossil forms. Establishing whether or not the loss of a specific bone is comparable across the phylogeny is, therefore, quite complicated. Another limitation is that the presence/absence of these skull bones are often used as characters in phylogenetic analyses based on morphological data. Consequently, mapping the presence/absence of these bones on phylogeny to study their evolution is somewhat circular. Nonetheless, quantifying patterns of loss and regain can still provide helpful insights into the comparable evolvability of certain bones, and whether convergent bone loss (even if by fusion) occurred in a modular manner (i.e. bones are lost together and not independently).

In light of this, we present the first reconstruction of the evolutionary history of bone loss in the tetrapod skull. We aim to set a framework for future discussions and to detect if there are distinct patterns of presence/absence, or loss/regain of bones that have not been recognised in previous studies due to more limited taxonomic samples or focus on specific groups.

The skull morphology of 313 extinct and extant tetrapod taxa were examined. The material includes finned and limbed stem-tetrapods, as well as extant tetrapods (lissamphibians plus amniotes) and their closest extinct relatives (Figs 2, 3; Suppl. material 1). The species sampling was designed to represent the phylogenetic and morphological diversity within tetrapods. We have focused solely on bones in the median and temporal series in the dermatocranium for two reasons: (1) the dorsal bones of the dermatocranium are typically the best-preserved skull bones in the fossil record, and (2) the median and temporal series represent the most and least stable series of the dermatocranium, respectively. The median series is largely involved with protecting the brain and other sensory organs, it includes the nasal, frontal, parietal and postparietal (sometimes referred to as the interparietal (Koyabu et al. 2012)). The temporal series, on the other hand, is variably made of the intertemporal, supratemporal, and tabular. The supratemporal and tabular are associated with the otic region of the skull (Clack et al. 2003), and the tabular along with the postparietal act as attachment sites for muscles associated with the neck (Carstens 2023).

Figure 2. 

Maximum likelihood ancestral state reconstruction of the presence/absence of the postparietal, intertemporal, supratemporal, and tabular (see key for colour associations) under Temnospondyl Hypothesis a. At the nodes, a filled in quadrant = presence; empty quadrant = absence; striped quadrant = presence/absence is ambiguous. On the branches, empty rectangle = loss of element; filled rectangle = gain of element. At the branch tips, filled in circle = presence of element; empty circle = absence of element; half-filled circle = mixed presence/absence of element. Tree A = Tetrapodomorpha, Temnospondyli, and Lissamphibia. Tree B = Seymouriamorpha, Diadectomorpha, Synapsida, and Lepospondyli. Tree C = Captorhinidae, Protorthyrididae, Araeoscelidia, Varanopidae, Neodiapsida, and Parareptilia.

Figure 3. 

Maximum likelihood ancestral state reconstruction of the presence/absence of the postparietal, intertemporal, supratemporal, and tabular (see key for colour associations) under Lepospondyl Hypothesis a. At the nodes, a filled in quadrant = presence; empty quadrant = absence; striped quadrant = presence/absence is ambiguous. On the branches, empty rectangle = loss of element; filled rectangle = gain of element. At the branch tips, filled in circle = presence of element; empty circle = absence of element; half-filled circle = mixed presence/absence of element. Tree A = Tetrapodomorpha and Temnospondyli. Tree B = Seymouriamorpha, Diadectomorpha, Synapsida, Lepospondyli, and Lissamphibia. Tree C = Captorhinidae, Protorthyrididae, Araeoscelidia, Varanopidae, Neodiapsida, and Parareptilia.

Two composite trees were constructed by hand in Mesquite (Maddison and Maddison 2021) which showed the relationships of all major stem and crown tetrapod groups with Lissamphibia either nested within Temnospondyli or Lepospondyli (Figs 2, 3; Suppl. materials 2, 3). While the phylogenetic position of several major clades is not well clarified, we chose to only consider the alternative hypotheses on the origin of lissamphibians (i.e. Temnospondyl and Lepospondyl Hypotheses) because: (1) the position of Lissamphibia radically affects the phylogenetic structure of early tetrapod relationships, and (2) as mentioned previously, the use of ‘loss characters’ to construct the Temnospondyl and Lepospondyl Hypothesis phylogenies impacts the interpretation of the loss of cranial bones.

The backbone of the trees was taken from the comprehensive and highly cited (e.g. Anderson et al. 2008b; Pyron 2011; Pardo et al. 2017b; Marjanović and Laurin 2019) strict consensus tree found by Ruta and Coates (2007), the stem tetrapods, Nectridea, and Aistopoda sections of the tree was left unchanged. Other portions of the tree, however, were altered. The microsaur topology was taken from MacDougall et al. (2021); the temnospondyl branch was replaced with trees recovered by Schoch (2013b) and Schoch (2022). The lissamphibian topology was taken from Jones et al. (2022), while the branch leading to the Reptiliomorpha was replaced by the tree from Ford and Benson (2020). For our purposes, the neodiapsid section of the Ford and Benson (2020) tree was not exhaustive enough as it only includes six taxa, all of which are extinct. To address this, the neodiapsid portion was replaced with that from Simoes et al. (2018). The position of Testudinata shown in the Simoes et al. (2018) tree, however, was altered so that it was a member of Archelosauria, as proposed by Crawford et al. (2015). The placement of Saurosphargidae was also updated from that shown in Simoes et al. (2018) to follow the position in Wang et al. (2022) which places them as sister to Sauropterygia. Within therapsids, the topology recovered by Huttenlocker et al. (2021) was used as a backbone, while the topology from Fraser-King et al. (2019) was used for Biarmosuchia, Anomodontia, Gorgonopsia, and Dinocephalia, and the topologies from Huttenlocker (2009) and Wallace et al. (2019) were used for Therocephalia and Cynodontia, respectively.

The trees were later time calibrated in R using the ‘timePaleoPhy’ function in the ‘paleotree’ package (Bapst 2012), with ‘randres’ set to TRUE to allow polytomies to be randomly resolved and using the ‘minMax’ date treatment. ‘minMax’ was chosen over other treatments such as ‘firstLast’ because it incorporates uncertainty by randomly assigning an age to a species within the provided time data (Bapst and Wagner 2022). This is often necessary when dealing with the vertebrate fossil record as it is rarely possible to provide precise first and last appearance dates, especially given that so many of the clades in the composite tree are poorly sampled. The minimum and maximum ages of the first appearance datum (FADs) for each tip were provided for the time calibration. The FAD ages were taken from the age of the rock formations that the oldest known specimen of each tip or clade was found in. There were four exceptions where it was not possible to ascertain the age of a tip based on geology, in these cases ages were taken from palynological analyses (Clack et al. 2019), estimations from molecular clock methods (van Tuinen and Dyke 2004), time calibrated phylogenetic analyses (Joyce et al. 2013), and biogeographically calibrated phylogenies (Leavitt et al. 2007) (see Suppl. material 4).

Presence/absence data for the bones in the temporal and median series in the skull roof were gathered for each of the 313 tips in the composite tree and scored as a binary trait. Where possible, presence/absence data was gathered from first hand examination of specimens, but this was not feasible in the majority of cases, and so data was largely taken from the literature (see Suppl. material 1). A bone element was defined based on its sutures separating it from its neighbours, therefore, any reported embryonic bones were marked as absent, except for the homology hypothesis for Mammaliaformes mentioned below. In instances where there are mixed reports on the presence/absence of a given bone, such as the intertemporal in Greererpeton and the temnospondyl Micromelerpeton, the bone was coded as absent as its presence is rare and its identity uncertain. Birds and frogs, which have frontoparietals (although see Arnaout et al. (2022) as there is still some debate on the homology of the frontal in birds), were coded as having both the frontal and parietal.

To ensure that uncertainties on bone homology were taken into account, several alternative hypotheses were collated from the literature, and a separate dataset was constructed for each with alternative coding for the relevant taxa. These different hypotheses of homology are shown in Table 2 and were based on: (1) the temporal series of the Early Jurassic stem-caecilian, Eocaecilia, which has a single bone that is presumed to be either the supratemporal or tabular (Jenkins et al. 2007); (2) the single interparietal in mammals, whose embryological data suggests may be composed of paired postparietals fused to the tabulars, instead of just the paired postparietals (Koyabu et al. 2012); (3) the median series in extant birds, which is composed of two bones that are either the frontal and parietal, or the frontoparietal and postparietal (Maddin et al. 2016; Arnaout et al. 2022); (4) the temporal series of the Permian nectridean, Diplocaulus, which is composed of a single bone that is either the supratemporal or tabular (Olson et al. 1951; Carroll et al. 1998), and (5) the uncertainty in the presence of a supratemporal in the stem testudines Odontochelys and Proganochelys (Gaffney and Meeker 1983; Li et al. 2008; Scheyer et al. 2022).

Maximum likelihood ancestral state reconstructions were conducted for each bone, except for the nasals, frontals, and parietals, as these were found to be present in all taxa. Analyses were performed on the different datasets considering the different homology hypotheses, and alternatively under the temnospondyl and lepospondyl phylogenetic hypotheses for the origin of Lissamphibia. The reconstructions were performed in R using the ‘ace’ function from the ‘ape’ package (Paradis and Schliep 2019). The weights and AICs (Akaike Information Criterion) from ancestral state reconstructions using the ER (Equal Rates, i.e., all possible character state transitions occur at the same rate) and ARD (All Rates Different, i.e., all possible character state transitions can occur at a different rate) models were compared for each bone, and it was determined that the ARD model was the most appropriate to use for all the bones (see Suppl. material 5). Plots showing the ancestral state reconstructions with the ARD model for each bone under Homology Hypothesis A in both the Temnospondyl and Lepospondyl Hypothesis can be found in the supplementary, along with the data and R script used (Suppl. materials 6–23). To reconstruct the ancestral states in nodes (i.e. presence or absence of the bones), the ratio of the likelihoods of two estimates is used to gauge the level of support for the maximum likelihood estimate. A likelihood ratio of 7.4:1 is commonly employed as a rough minimum threshold to determine the likely state of a node (Schulter et al. 1997; Rüber et al. 2004; Dalerum 2007). With this threshold, gains and losses of bones were identified by manually tracing character changes in the branches. At each node the presence % likelihood returned from the analysis was multiplied by 7.4, if this value was less than the absence % likelihood at that node, then the bone in question was deemed to be absent. If a transition was thought to have occurred at a node (e.g. from present to absent), then a second calculation was undertaken to determine if the character change was statistically significant or not. A transition threshold was calculated by dividing the absence % likelihood by 7.4, if this value was equal to or greater than the presence % likelihood then the transition from present to absent was statistically significant.

In summary, the simplification of the dermatocranium is a well-documented yet poorly understood phenomenon which began in the Devonian and evolved parallelly in various tetrapod lineages into the Mesozoic. By conducting ancestral state reconstructions on the presence/absence of the median and temporal series, we have shown that the evolutionary loss of bones in the dermatocranium was much more complex than initially thought, and likely involved a mixture of developmental, ecological, and functional drivers. As no obvious correlation was found between skull simplification and lifestyle or body size, further work will be required to determine what these factors may be and how they interact with one another both in individual lineages, and across the evolutionary history of different tetrapod lineages. Our analyses also showed that the temporal series did not form an evolutionary module, and consequently that the developmental pathways influencing the presence/absence of these individual bones were unlikely to be strongly integrated as the loss of these bones is quite variable. In general though, the intertemporal is lost first, then followed by the supratemporal, then the tabular and/or postparietal. However, although most groups seem to follow this pattern of loss, this was not the case in Sauropsida (i.e. Eureptilia and Parareptilia) where the tabular or postparietal was the second bone to be lost and not the supratemporal. Of the bones studied, the supratemporal was the most evolvable given that it was lost and regained the most often, while the nasal, frontal, and parietal were the least evolvable as they were always present. Interestingly, the sequence in which the temporal and median series bones were lost did not reflect the order in which they ossify in ontogeny.

The analyses presented here demonstrate that the evolution of skull simplification is much more complex than previously realised. Further work is needed to fully capture the intricacies involved, and we would encourage future studies to focus on reassessing skull homology across tetrapodomorphs, and to apply embryological methods to determine the mode of bone loss in the skulls of extant groups.

The authors would like to thank Marcello Ruta and Torsten Scheyer for their kind and constructive reviews which helped improve the manuscript. KJK would like to thank Johannes Müller and Mark MacDougal for helpful discussions on amniotes, and parareptile and recumbirostran phylogeny respectively; Davide Foffa for providing help with R; Antoine Verrière for giving advice both on R and on the figures presented here; and Yara Haridy for proof reading. This work is part of KJK’s PhD project and was funded by the Deutsche Forschungsgemeinschaft (grant number 442217617) awarded to MD.