Non-technical Summary

Permo-Triassic synapsids are the ancestors of mammals. Study of the synapsid brain is 150 years old and is reviewed here. Some evidence supports that mammalian ancestors have been capable of complex behavior such as parental care, brooding, intraspecific combat, display, and gregariousness for up to 265–260 million years. The study of the brain and sense organs supports that they evolved hair and warm blood 240–230 million years ago. Brain evolution was driven by a neurosensory revolution during the Triassic as terrestrial ecosystems began to be ruled by archosaurs.

A brief history of synapsid paleoneurology

Brain evolution is traced through the fossil record by studying the internal cast of the brain cavity of fossilized skulls, usually referred to, more simply, as the endocranial cast or endocast (Jerison, Reference Jerison1973; de Sousa et al., Reference de Sousa, Beaudet, Calvey, Bardo and Benoit2023; Rowe, Reference Rowe, Dozo, Paulina-Carabajal, Macrini and Walsh2023). Study of the paleoneurology of synapsids began 150 years ago, in 1876, with the publication of the endocast of the cynodont grade (i.e., non-mammalian cynodonts; Fig. 1) species Nythosaurus larvatus Owen, Reference Owen1876, by Richard Owen (Fig. 2.1) (Owen, Reference Owen1876; Pusch et al., Reference Pusch, Kammerer, Fernandez and Fröbisch2022). For almost a hundred years, to access data on the synapsid nervous system, paleoneurologists had to rely on the slim chances of preservation of natural endocasts of neurosensory cavities and canals (e.g., Watson, Reference Watson1913; Cox and Broom, Reference Cox and Broom1962), complete preparation (e.g., Case, Reference Case1914; Tatarinov, Reference Tatarinov1965; Kemp, Reference Kemp1969, Reference Kemp1979), as well as on destructive methods such as sectioning, serial sectioning, and serial grinding (e.g., Sollas and Sollas, Reference Sollas and Sollas1913; Olson, Reference Olson1944; Boonstra, Reference Boonstra1968; see Benoit and Jasinoski, Reference Benoit and Jasinoski2016, for a review). In a rather unusual manner for the field, paleoneurologists studying synapsids have mostly relied upon natural endocasts or digital 3D reconstructions, but seldom made artificial endocasts using latex or plaster of Paris (Edinger, Reference Edinger1975).

Figure 2. Endocasts, in dorsal view, illustrating the history of techniques used in synapsid paleoneurology. (1) Natural endocast of Nythosaurus larvatus described by Owen (Reference Owen1876), NHMUK PV R 1715 from the Natural History Museum UK, courtesy of M. Day; (2) first digital endocast based on the CT-scanned skull of a Thrinaxodon liorhinus Seeley, Reference Seeley1894, published by Rowe et al. (Reference Rowe, Carlson, Bottorff and Olson1995) as a Compact Disc; (3) digital endocast based on synchrotron data of Thrinaxodon liorhinus by Fernandez et al. (Reference Fernandez, Abdala, Carlson, Cook, Rubidge, Yates and Tafforeau2013). Endocasts not to scale.

For most of the twentieth century, data were scarce and progress was slow, yet steady, mostly thanks to the noted contributions of Harry J. Jerison (Reference Jerison1973), who introduced the Encephalization Quotient as a measure of relative brain size, Zofia Kielan-Jaworowska for her work on Mesozoic mammal paleoneurology (Kielan-Jaworowska, Reference Kielan-Jaworowska1983, Reference Kielan-Jaworowska1984), and Juan C. Quiroga, who described the endocast of many non-mammaliaform cynodonts in the late 1970s and early 1980s (see Jerison, Reference Jerison1973; Edinger, Reference Edinger1975; Hopson et al., Reference Hopson, Gans, Northcutt, Ulinski, Gans, Northcutt and Ulinski1979; Kielan-Jaworowska et al., Reference Kielan-Jaworowska, Cifelli and Luo2004; Kerber et al., Reference Kerber, Roese-Miron, Bubadué and Martinelli2024a, for reviews).

Progress was slowed, in large part, by lack of ossification of the synapsid braincase, particularly on its ventral and lateral aspects (Kielan-Jaworowska et al., Reference Kielan-Jaworowska, Cifelli and Luo2004; Kemp, Reference Kemp2009; Benoit et al., Reference Benoit, Jasinoski, Fernandez and Abdala2017a). Early synapsids, with a few exceptions such as the dinocephalians, biarmosuchians, and Kawingasaurus (Benoit et al., Reference Benoit, Fernandez, Manger and Rubidge2017b; Laaß and Kaestner, Reference Laaß and Kaestner2017), lack a laterally expanded epipterygoid and an ossified cribriform plate anteriorly. Dorsally, the internal surface of the supraoccipital is excavated by a deep unossified zone (like in modern turtles, in which it is filled up by cartilage; Werneburg et al., Reference Werneburg, Evers and Ferreira2021), and ventrally, below the olfactory bulbs, the more or less ossified orbitosphenoids are floating as part of the mostly cartilaginous median septum (Benoit et al., Reference Benoit, Jasinoski, Fernandez and Abdala2017a).

The process of ossification of the braincase began in probainognathian cynodonts only and was achieved in early Mammaliaformes (Kielan-Jaworowska et al., Reference Kielan-Jaworowska, Cifelli and Luo2004; Benoit et al., Reference Benoit, Jasinoski, Fernandez and Abdala2017a; Norton et al., Reference Norton, Abdala and Benoit2023). The ventral flexure of the brain (Benoit et al., Reference Benoit, Fernandez, Manger and Rubidge2017b) partly solved the long-lasting question of the ventral limits of the endocast in early synapsids (Kemp, Reference Kemp2009), but the lack of ossified walls and floor still poses major challenges to reconstructing the endocast, nerves, and blood vessels in many synapsid taxa (e.g., Laaß et al., Reference Laaß, Schillinger and Kaestner2017a; Bazzana et al., Reference Bazzana, Evans, Bevitt and Reisz2022; Bazzana-Adams et al., Reference Bazzana-Adams, Evans and Reisz2023).

Synapsid paleoneurology was revived with the application of new imaging techniques such as microcomputed tomography (CT) scanning using X-ray, synchrotron, or neutron beams to create a series of digital slices of fossilized skulls. These images are processed in silico to extract the 3D volume of the normally out-of-reach brain cavity (endocast), inner ear (bony labyrinth), and other nervous structures (such as the maxillary canal of the trigeminal nerve) through a process called segmentation. The first CT-scanned synapsid was a Thrinaxodon skull in 1993 (Fig. 2.2), which was shortly preceded by the CT scanning of the skull of the early mammaliaform Morganucodon (Luo and Ketten, Reference Luo and Ketten1991; Rowe et al., Reference Rowe, Carlson, Bottorff and Olson1995; Rowe, Reference Rowe1996). The subsequent spread of CT scanning (mostly X-ray-based CT scanning) in paleontology labs globally provided access to an unprecedented quantity of synapsid, early mammaliaform, and Mesozoic mammal endocranial data (e.g., Hurum, Reference Hurum1998; Macrini, Reference Macrini2006; Macrini et al., Reference Macrini, Rougier and Rowe2007; Luo et al., Reference Luo, Ruf, Schultz and Martin2011; Rowe et al., Reference Rowe, Macrini and Luo2011; Rodrigues et al., Reference Rodrigues, Ruf and Schultz2014; Laaß, Reference Laaß2015a, Reference Laaßb; Maier and Ruf, Reference Maier and Ruf2016; Benoit et al., Reference Benoit, Fernandez, Manger and Rubidge2017b, Reference Benoit, Manger, Fernandez and Rubidgec; Laaß and Kaestner, Reference Laaß and Kaestner2017; Laaß et al., Reference Laaß, Schillinger and Kaestner2017a; Rodrigues et al., Reference Rodrigues, Martinelli, Schultz, Corfe, Gill, Soares and Rayfield2018; Pavanatto et al., Reference Pavanatto, Kerber and Dias‐da‐Silva2019; Hoffmann et al., Reference Hoffmann, Rodrigues, Soares and Andrade2021; Araújo et al., Reference Araújo, Macungo, Fernandez, Chindebvu and Jacobs2022a; Bazzana-Adams et al., Reference Bazzana-Adams, Evans and Reisz2023; Rowe, Reference Rowe, Dozo, Paulina-Carabajal, Macrini and Walsh2023; Kerber et al., Reference Kerber, Roese-Miron, Bubadué and Martinelli2024a; Pusch et al., Reference Pusch, Kammerer and Fröbisch2024; Medina et al., Reference Medina, Martinelli, Gaetano, Roese-Miron, Tartaglione, Backs, Novas and Kerber2025, and references therein), including studies of the inner ear (Luo and Ketten, Reference Luo and Ketten1991; Luo et al., Reference Luo, Ruf, Schultz and Martin2011; Rodrigues et al., Reference Rodrigues, Ruf and Schultz2014; Laaß, Reference Laaß2015a, Reference Laaßb; Benoit et al., Reference Benoit, Manger, Fernandez and Rubidge2017c; Araújo et al., Reference Araújo, David, Benoit, Lungmus and Stoessel2022b; Bazzana et al., Reference Bazzana, Evans, Bevitt and Reisz2022; Bazzana-Adams et al., Reference Bazzana-Adams, Evans and Reisz2023) and trigeminal nerves (Benoit et al., Reference Benoit, Manger and Rubidge2016a, Reference Benoit, Manger, Fernandez and Rubidgeb, Reference Benoit, Abdala, Manger and Rubidgec, Reference Benoit, Angielczyk, Miyamae, Manger, Fernandez and Rubidge2018, Reference Benoit, Ruf, Miyamae, Fernandez, Rodrigues and Rubidge2020a, Reference Benoit, Legendre, Farke, Neenan, Mennecart, Costeur, Merigeaud and Mangerb, Reference Benoit, Norton and Jirah2023a; Laaß and Kaestner, Reference Laaß and Kaestner2017; Wallace et al., Reference Wallace, Martínez and Rowe2019; Muchlinski et al., Reference Muchlinski, Wible, Corfe, Sullivan and Grant2020; Duhamel et al., Reference Duhamel, Benoit, Rubidge and Liu2021; Bazzana et al., Reference Bazzana, Evans, Bevitt and Reisz2023; Norton et al., Reference Norton, Abdala and Benoit2023; Fonseca et al., Reference Fonseca, Martinelli, Gill, Rayfield, Schultz, Kerber, Ribeiro and Soares2024a; Miyamae et al., Reference Miyamae, Benoit, Ruf, Sibiya and Bhullar2024; Pusch et al., Reference Pusch, Kammerer and Fröbisch2024).

Subsequent synchrotron (Fig. 2.3) and neutron imaging enabled data extraction from the most challenging, largest, or most metallic nodule-rich specimens (Laaß et al., Reference Laaß, Schillinger and Werneburg2017b; von der Heyden et al., Reference von der Heyden, Benoit, Fernandez and Roychoudhury2020) so that the dataset of synapsid endocasts has now grown to almost 60 genera (Table 1).

Table 1. Dataset of Body Mass (BM) and Endocast Volume (EV), and volume of the Olfactory Bulbs (OB) of all synapsids, early mammaliaforms, and Mesozoic mammals reported in the literature, with sources and specimen numbers. Notes: 1, Pineal tube excluded; 2, Mistakenly identified as Moschops capensis and numbered AM6556; 3, OB calculated from provided measurements (assuming they have an ellipsoid shape); 5, BM from skull length (Edinger, Reference Edinger1955); 6, Value marked with a * is probably wrong; 7, EV calculated from figures using graphic double integration (Jerison, Reference Jerison1973); 8, BM from skull length (Abdala, Reference Abdala2007); 9, Mistakenly identified as Thrinaxodon and numbered BMHR1713; 10, BM from skull length (Watson, Reference Watson1913); 11, Initially identified as Probelesodon kitchingi; 12, Initially identified as cf. Probelesodon; 13, Initially identified as Chiniquodon sp.; 14, Initially identified as Probelesodon sp.; 15, Initially identified as Brailistherium riograndensis; 16, EV calculated back from provided EQ and BM. See Appendix 1 for authorship of named species

Parental care, brooding, and burrowing

Life in groups and parental care are two of the most recognizable mammalian behavioral traits and can be traced back to as far as the late Carboniferous, in pelycosaurs (Botha-Brink and Modesto, Reference Botha-Brink and Modesto2007; Spindler, Reference Spindler2015; Maddin et al., Reference Maddin, Mann and Hebert2020). Fascinatingly, the oldest unambiguously documented social behavior among Paleozoic tetrapods is found in synapsids and consists of evidence of adult–juvenile aggregations (Maddin et al., Reference Maddin, Mann and Hebert2020). Synapsids have the richest fossil record of adult–juvenile aggregations among tetrapods, as dozens of occurrences have been found globally across pelycosaurs, therapsids, and cynodonts (Bandyopadhyay and Southwood, Reference Bandyopadhyay and Southwood1997; Botha-Brink and Modesto, Reference Botha-Brink and Modesto2007; Spindler, Reference Spindler2015; Jasinoski and Abdala, Reference Jasinoski and Abdala2017; Hoffman and Rowe, Reference Hoffman and Rowe2018; Benoit, Reference Benoit2019; Smith et al., Reference Smith, Angielczyk, Benoit and Fernandez2021). In particular, dicynodont trackways of juveniles walking alongside adults (Smith, Reference Smith1993) and the exceptional preservation of at least 23 adults and juveniles found together strongly supports parental care and life in herds in this clade (Bandyopadhyay and Southwood, Reference Bandyopadhyay and Southwood1997).

Recent discoveries suggest that burrows may have been used as brooding chambers in early synapsids (Smith et al., Reference Smith, Angielczyk, Benoit and Fernandez2021). Burrowing, which provides shelter during long episodes of torpor, aestivating, and hibernating, evolved in synapsids during the end-Capitanian extinction (Smith et al., Reference Smith, Angielczyk, Benoit and Fernandez2021; Marchetti et al., Reference Marchetti, MacDougall, Buchwitz, Canoville, Herde, Kammerer and Fröbisch2024), while tetrapod lineages were becoming extinct in larger numbers than they would later during the end-Permian “Great Dying” (Day et al., Reference Day, Ramezani, Bowring, Sadler, Erwin, Abdala and Rubidge2015; Day and Rubidge, Reference Day and Rubidge2021). This supports that burrowing gave synapsids an advantage during mass-extinction events (Fernandez et al., Reference Fernandez, Abdala, Carlson, Cook, Rubidge, Yates and Tafforeau2013; Botha-Brink, Reference Botha-Brink2017; Whitney and Sidor, Reference Whitney and Sidor2020). Coincidentally, using burrows as brooding chambers likely contributed to tightening the bonds between parents and juveniles at the very root of the mammalian lineage (Smith et al., Reference Smith, Angielczyk, Benoit and Fernandez2021).

In contrast to the extensive evidence of possible parental care, a large aggregation of one adult and 38 neonates of one of the closest relatives of mammals among cynodonts, the tritylodontid Kayentatherium, was described by Hoffman and Rowe (Reference Hoffman and Rowe2018). Such large litter size is more consistent with a reptilian-like reproductive biology than with a mammalian one. Coupled with the presence of erupted teeth in neonates, this supports that newborn cynodonts were self-sufficient and needed no milk and little parental care from the adults (Hoffman and Rowe, Reference Hoffman and Rowe2018; Benoit, Reference Benoit2019). In addition, litter size has been shown to decrease rapidly after birth, which is also consistent with scarce protection and food supply from parents (Benoit, Reference Benoit2019). To reconcile both lines of evidence, it has been proposed that juvenile cynodonts were engaging in parasitism or opportunistically taking advantage of the adults, rather than enjoying reciprocal, mammal-like parental care (Benoit, Reference Benoit2019). The possibility that parasitism evolutionarily preceded parental care in the lineage leading to mammals is intriguing but will require more fossil data to be addressed.

Head-butting and gregariousness

Besides parental care, head-butting is the second best studied and supported paleo-behavior in synapsids, mostly in the dinocephalians (Woodruff and Ackermans, Reference Woodruff and Ackermans2024). This was originally suggested by Brink (Reference Brink1958) and studied more in depth by later authors (Barghusen, Reference Barghusen1975; Benoit et al., Reference Benoit, Manger, Fernandez and Rubidge2016b, Reference Benoit, Manger, Norton, Fernandez and Rubidge2017e, Reference Benoit, Kruger, Jirah, Fernandez and Rubidge2021c, Reference Benoit, Dollman, Smith and Manger2023b, Reference Benoit2024; Benoit and Midzuk, Reference Benoit and Midzuk2024; Bolton and Benoit, Reference Benoit2024). In extant species, intraspecific agonistic behaviors (combat and display) are mostly used to secure mates and territory, and ensure dominance, implying the existence of a stratified society subjected to some form of hierarchical ranking within social groups (Geist, Reference Geist1966; Emlen, Reference Emlen2008). Head-butting implies that some synapsids may have been social animals more than 160 million years before the earliest evidence of such behavior in mammals (Weaver et al., Reference Weaver, Varricchio, Sargis, Chen, Freimuth and Wilson Mantilla2021). Parallel trackways (de Klerk, Reference de Klerk2002), communal latrines (Fiorelli et al., Reference Fiorelli, Ezcurra, Hechenleitner, Argañaraz, Taborda, Trotteyn, von Baczko and Desojo2013), and the rich record of aggregations of individuals that fossilized together in pelycosaurs (Botha-Brink and Modesto, Reference Botha-Brink and Modesto2007; Spindler, Reference Spindler2015), dinocephalians (Benoit et al., Reference Benoit, Norton and Jirah2023a), dicynodonts (Bandyopadhyay and Southwood, Reference Bandyopadhyay and Southwood1997; Viglietti et al., Reference Viglietti, Smith and Compton2013), and cynodonts (Jasinoski and Abdala, Reference Jasinoski and Abdala2017; Figueiredo et al., Reference Figueiredo, Melo, Neto, da Rosa and Pinheiro2024) also supports the theory that life in groups, some reaching up to 46 individuals, was a widespread behavior.

Dinocephalian adaptations to head-butting (Fig. 3.1) open a window into ancient synapsid behavior and life in groups. These adaptations include (1) thickening of cranial bones (pachyostosis), chiefly the skull roof and braincase walls (orbitosphenoid and epipterygoids), to protect the brain, and absorb and transfer the energy of the impact towards the occipital condyles; (2) the multi-layered skull roof, which is made of three layers of bone tissue (a surface layer of dense cortical bone, an intermediate layer of radially oriented cells, and a deep layer of spongy bone) and is similar to that of pachycephalosaurid dinosaurs (Benoit et al., Reference Benoit, Manger, Fernandez and Rubidge2016b, Reference Benoit, Manger, Norton, Fernandez and Rubidge2017e; Woodruff and Ackermans, Reference Woodruff and Ackermans2024); (3) thickened non-neural tissue (adnexa) surrounding the brain for cushioning, which manifests on the endocast as a distinct excess of encephalization in head-butting dinocephalians (Benoit and Midzuk, Reference Benoit and Midzuk2024); (4) re-orientation of the braincase, which is tilted backwards to move the parietal foramen away from the fighting surface (Fig. 4.1, 4.2); (5) re-orientation of the occiput to align the fighting surface, occipital condyles, and neck vertebrae to help dissipate the energy of the impact into the body (Fig. 3.1); and (6) the presence of a tilting of the plane of the lateral semicircular canal with respect to the skull (Fig. 4.1), which, in modern mammals, is significantly correlated with the practice of head-butting behavior (Benoit et al., Reference Benoit, Legendre, Farke, Neenan, Mennecart, Costeur, Merigeaud and Manger2020b). All these adaptations are present in modern head-butting taxa but are exaggerated in the tapinocephalids, which were seemingly the dinocephalian taxa that were the best adapted to this behavior (Benoit et al., Reference Benoit, Manger, Norton, Fernandez and Rubidge2017e, Reference Benoit, Kruger, Jirah, Fernandez and Rubidge2021c, Reference Benoit, Dollman, Smith and Manger2023b; Benoit and Midzuk, Reference Benoit and Midzuk2024; Woodruff and Ackermans, Reference Woodruff and Ackermans2024).

Figure 3. Short- and long-term consequences of head butting in tapinocephalid dinocephalians. (1) Routes through the braincase and to the neck taken by the energy of a head-on impact, around the endocast (left) and on the surface of the skull (right). (2) Distribution of cranial pathologies in tapinocephalid dinocephalians (left) compared to the stress distribution resulting from the finite element analysis (FEA, high stress in red) simulation of head butting (top, lateral view; bottom, dorsal views). (3) Digital cross-section through the braincase of a Moschognathus (AM 4950) showing the presence of an abscess (left) and two magnifications of the abscess area (right). (4) Interpretive drawing of (3) comparing the position of the abscess (left side of the skull) to the route taken by a head-on impact on the fighting surface (right side of the skull). Abbreviations: Abs = abscess; Bas = basicranium; Heal = healing bone tissue; Drain = pus drainage canal; FS = fighting surface; Osp = orbitosphenoid. Red arrows indicate the route taken by the energy of the impact. (1) and (2) not to scale.

Figure 4. Illustration of the hypothesized trade-off between the size of the canine and cranial adaptations to head-butting in mid-Permian dinocephalians (1), the skull of modern ruminants (2), and pictures of modern cervids (3). (1, 2) Orientation of the braincase (marked by the orientation of the lateral semicircular canal of the inner ear) in dinocephalians (1) from left to right: Anteosaurus, Moschognathus, and a derived tapinocephalid indet., and modern ruminants (2) from left to right: Moschus, Muntiacus, and Connochaetes; transparent skulls aligned on the plane of their lateral semicircular canal (white dashed line). (3) Pictures of living cervids in lateral view (from left to right: Hydropotes, Muntiacus, and Cervus). LSC = lateral semicircular canal. Not to scale.

The head-butting capability of dinocephalians was confirmed by using 3D digital finite element analyses (FEAs), to test and compare the ability of their skulls to withstand such impacts (Bolton and Benoit, Reference Bolton and Benoit2024). The genera considered are Struthiocephalus and Moschops (Tapinocephalidae), Estemmenosuchus (Estemmenosuchidae), Anteosaurus (Anteosauria), and Jonkeria (Titanosuchidae). These models were compared in terms of FEA stress distributions to Stegoceras, a pachycephalosaurid dinosaur considered to be suitably adapted for head-butting, performing better in FEAs than known high-impact head-butting ungulates (Snively and Theodor, Reference Snively and Theodor2011), and Hipposaurus, a basal biarmosuchian that has none of the typical adaptations for head-butting (Benoit et al., Reference Benoit, Araujo, Lund, Bolton, Lafferty, Macungo and Fernandez2024).

The dinocephalian models were created by Midzuk (Reference Midzuk2020), adjusted by Bolton (Reference Bolton2024), and analyzed with ANSYS Workbench Software using two load protocols: one for head-on head-butting and the other applied onto the supra-orbital boss to reconstruct flank-butting (Bolton, Reference Bolton2024; Bolton and Benoit, Reference Bolton and Benoit2024). The load magnitudes applied were 1000 N and 3400 N, which conforms to previous literature (Farke, Reference Farke2008; Snively and Cox; Reference Snively and Cox2008; Snively and Theodor, Reference Snively and Theodor2011). The dinocephalian skulls outperformed both the low-suitability control of Hipposaurus and the high-suitability control of Stegoceras, with lower stress levels and less pervasive stress distributions, in all protocols (Bolton, Reference Bolton2024; Bolton and Benoit, Reference Bolton and Benoit2024). In terms of general FEA performances, the Struthiocephalus and Moschops models fared the best, which is in line with earlier observations that tapinocephalid dinocephalians appear to be better adapted for head-butting than non-tapinocephalids (Barghusen, Reference Barghusen1975; Benoit et al., Reference Benoit, Kruger, Jirah, Fernandez and Rubidge2021c). The FEAs support that the extreme pachyostosis and cranial adaptations of dinocephalians result in extraordinary mechanical resilience. Even Estemmenosuchus, the smallest dinocephalian skull model by volume, with the worst FEA performance of the group, performed better than the standard model of Stegoceras (Bolton, Reference Bolton2024; Bolton and Benoit, Reference Bolton and Benoit2024).

These results were validated by searching for pathologies in 202 dinocephalian skulls (Bolton, Reference Bolton2024; Bolton and Benoit, Reference Bolton and Benoit2024). Of these, six specimens were observed to have external, healed pathologies, and one was observed with an internal pathology (Fig. 3.1–3.3). The pathologies were found in the regions surrounding the orbit and braincase, which corresponds to the most stressed regions of the skull in the FEA results (Fig. 3.2). No pathologies appear in low-stress regions such as the snout and lower jaw. So not only was the dinocephalian skull mechanically capable of withstanding head-butting, as supported by the FEAs, but the distributions of simulated stress coincide with those of cranial pathologies, which provides further evidence that their heads were used as weapons.

The healed or partly healed nature of the injuries found on tapinocephalid dinocephalian skulls indicates that the animals recovered from these collisions. This indicates that the combats were non-lethal, which is an expected outcome of ritualized intraspecific agonistic behavior (Geist, Reference Geist1966). One of these pathologies was observed within the skull of a sub-adult dinocephalian as a result of synchrotron scanning (Benoit et al., Reference Benoit, Araujo, Lund, Bolton, Lafferty, Macungo and Fernandez2024). It is located deep inside the skull at the suture between the orbitosphenoid, frontal, and postorbital bones. The healed bone tissue has a distinct histology that can be observed far anteriorly in the frontal sinus. Drainage canals to evacuate the pus into the sinuses and orbit are evident (Fig. 3.2, 3.3). The location of this pathology excludes that it is the result of a bite mark. It is exactly on the pathway that the energy of head-butting would have traveled through to exit the skull (Fig. 3.4), which supports that it is the result of head-butting. The still incompletely ossified braincase of this non-adult individual likely would have been compressible enough to enable the orbitosphenoid to puncture the frontal bone upon impact, thus causing the injury (Benoit et al., Reference Benoit, Araujo, Lund, Bolton, Lafferty, Macungo and Fernandez2024).

The occurrence of this pathology in a non-adult individual suggests that it may have been the result of play-fighting, which is a well-documented behavior among juveniles of social, head-butting ungulate species (Hass and Jenni, Reference Hass and Jenni1993; Pellis and Pellis, Reference Pellis, Pellis, Bekoff and Byers1998, Reference Pellis and Pellis2017). This behavior is not documented in solitary mammals or reptiles, except tyrannosaurids (Peterson et al., Reference Peterson, Henderson, Scherer and Vittore2009). If confirmed in dinocephalians, play-fighting would imply prolonged time spent in groups (at least among juveniles), and imply refined, ritualized social signaling to convey the intention to play without aggression (Hass and Jenni, Reference Hass and Jenni1993; Pellis and Pellis, Reference Pellis, Pellis, Bekoff and Byers1998, Reference Pellis and Pellis2017). This pathology may thus be the most compelling piece of evidence for year-long social interactions in synapsids (i.e., not just during the mating season, and not in adults only), but more examples like this one must be found to bolster these conclusions.

Although it remains speculative at this point, it is noticeable that the vast range of morphological variations in the cranial adornments of dinocephalians may be the reflection of an equally diverse array of agonistic behavior (Bolton, Reference Bolton2024; Bolton and Benoit, Reference Benoit2024), for example: enlarged caniniforms for face biting in Anteosaurus and Jonkeria, a pachyostotic cranial dome for ramming in Moschops, Ulemosaurus, Tapinocaninus, and Moschognathus, horns on the nasal, tabular, and zygomatic bones for giraffe-like flank-butting in Styracocephalus, complex antler-like adornments for locking and pushing in Estemmenosuchus, and a single frontal horn for stabbing in Struthiocephalus (Fig. 5). This suggests a behavioral repertoire of fighting styles in dinocephalians that would have been comparable to that of modern ruminants (Vander Linden and Dumont, Reference Vander Linden and Dumont2019; Woodruff and Ackermans, Reference Woodruff and Ackermans2024), with the most potent weapons likely being found in the more solitary species whereas the less dangerous adornments would have occurred in the more gregarious ones, as in modern ungulates (Fig. 4; Geist, Reference Geist1966; Emlen, Reference Emlen2008; Cabrera and Stankowich, Reference Cabrera and Stankowich2020).

Figure 5. The diversity of possible fighting styles in dinocephalians. (1) Face biting in Anteosaurus; (2) head butting and ramming in Moschops; (3) flank butting in Styracocephalus; (4) locking and pushing in Estemmenosuchus; (5), stabbing in Struthiocephalus (position of a keratinous horn reconstructed as dotted lines).

Display behavior

Display (interspecific, sexual, or agonistic) is one of the most conspicuous forms of behavior an animal can produce. Yet it leaves very little trace in the fossil record, making it the most speculative behavior to trace through evolution (Boucot and Poinar, Reference Boucot and Poinar2010; Lockley et al., Reference Lockley, McCrea, Buckley, Lim and Matthews2016). Despite these challenges, there is evidence to support that synapsids may have engaged in such behavior. The large and slightly dimorphic sail on the back of some early Permian Dimetrodon limbatus Cope, Reference Cope1877, may have been the earliest synapsid display organ because models suggest it would have played a minor role in thermoregulation (Tomkins et al., Reference Tomkins, LeBas, Witton, Martill and Humphries2010; Rega et al., Reference Rega, Noriega, Sumida, Huttenlocker, Lee and Kennedy2012). Caniniform teeth that are larger than other marginal teeth have been present in synapsids since the origin of the clade (Angielczyk and Kammerer, Reference Angielczyk, Kammerer, Zachos and Asher2018) and evolved into conspicuous saber-like teeth in gorgonopsians, therocephalians, and cynodonts, into tusks in the herbivorous dicynodonts, and, eventually, into true canines in modern mammals (Benoit et al., Reference Benoit, Manger, Fernandez and Rubidge2016b). There are three main lines of argument to support that these caniniform teeth were presumably not only used for food acquisition but were also used as tools for social interaction in early synapsids.

Firstly, healed bite marks have been found on the fossilized skulls of some synapsids, and, in at least one taxon, could be safely identified as having been made by conspecifics (Benoit et al., Reference Benoit, Browning and Norton2021a, Reference Benoit, Dollman, Smith and Manger2023b). This is a direct indication of non-lethal face-biting, an intraspecific agonistic behavior that is commonly encountered in modern mammals. Secondly, enlarged caniniforms are not only observed among carnivorous taxa. Most herbivorous synapsids possess distinct caniniforms or tusks too, including the anomodont Tiarajudens, most dicynodonts, the titanosuchid dinocephalians, and cynognathian cynodonts (Cisneros et al., Reference Cisneros, Abdala, Jashashvili, de Oliveira Bueno and Dentzien-Dias2015; Benoit et al., Reference Benoit, Manger, Fernandez and Rubidge2016b, Reference Benoit, Browning and Norton2021a). Although wear facets preserved in some dicynodont specimens suggest the tusks may have been used for grubbing (King, Reference King1990), the presence of exaggeratedly large and ever-growing caniniforms in herbivores is inconsistent with them being used solely for food acquisition (Geist, Reference Geist1966, Reference Geist1972). Thirdly, the synapsids that lost their enlarged caniniforms evolved other forms of cranial adornments, such as bosses and horns, including the tapinocephalid dinocephalians, Cistecephalus, or Pelanomodon (Geist, Reference Geist1972; Nasterlack et al., Reference Nasterlack, Canoville and Chinsamy2012; Benoit et al., Reference Benoit, Manger, Fernandez and Rubidge2016b, Reference Benoit, Kruger, Jirah, Fernandez and Rubidge2021c, Reference Benoit2024; Kammerer et al., Reference Kammerer, Angielczyk and Fröbisch2016). This is similar to the evolutionary trade-off observed in ungulate mammals, such as brontotheres, rhinocerotids, and ruminants (Geist, Reference Geist1966; Stanley, Reference Stanley1974; Emlen, Reference Emlen2008; Cabrera and Stankowich, Reference Cabrera and Stankowich2020), in which the size of the horns and antlers used for fighting and display is inversely proportional to that of the canines or tusks (Fig. 4). The fact that reduction of the former would be compensated for by development of the latter supports the hypothesis that caniniform teeth and cranial adornments had the same functions in both combats and displays in synapsids. Finally, these enlarged caniniform teeth (or corresponding processes) are often sexually dimorphic in synapsids (Sullivan et al., Reference Sullivan, Reisz and Smith2003; Kammerer et al., Reference Kammerer, Angielczyk and Fröbisch2011, Reference Kammerer, Angielczyk and Fröbisch2016; Nasterlack et al., Reference Nasterlack, Canoville and Chinsamy2012; Cisneros et al., Reference Cisneros, Abdala, Jashashvili, de Oliveira Bueno and Dentzien-Dias2015; Benoit et al., Reference Benoit, Manger, Fernandez and Rubidge2016b; Pinto et al., Reference Pinto, Marshall, Nesbitt and Varajão de Latorre2024), indicating that they likely evolved either for sexual display or intraspecific fights, or both, as in modern ungulate mammals (Geist, Reference Geist1966; Emlen, Reference Emlen2008).

The practice of sexual display in early synapsids is further supported by the CT-assisted study of the maxillary bosses in two synapsid genera: Choerosaurus and Pachydectes (Benoit et al., Reference Benoit, Manger, Fernandez and Rubidge2016b, Reference Benoit2024). The maxillary bosses on the sides of the skull of these two taxa were likely expensive to grow. Growing and maintaining these large, and pachyostotic cranial adornments would have been costly, which supports that they improved chances of reproduction through protection against predators, sexual display, intraspecific combat, or for species recognition in large sympatric herds. Protection against predators would be redundant with the presence of an enlarged caniniform, which both taxa retained. Species recognition in large herds is unlikely as neither biarmosuchians nor therocephalians have been recovered in aggregations (see above). These bosses occur in animals that have a rather lightly built skull, which could not have withstood the use of these bosses for combat. Had the bosses been used to smash sideways, the maxilla would have collapsed inwards under the impact due to the absence of a secondary palate in both Choerosaurus and Pachydectes (Benoit et al., Reference Benoit, Manger, Fernandez and Rubidge2016b, Reference Benoit2024). Moreover, in both taxa, the bosses are full of canals for delicate nerves and blood vessels that would have been easily damaged upon impact (Fig. 6). In contrast, these nerves and blood vessels are consistent with the possible presence of a keratinous pad covering the boss and capable of blushing quickly on command thanks to the rich blood supply and fine nervous control they provided. As such, these structures are better interpreted as having been used for a more ritualized purpose, such as display (Benoit et al., Reference Benoit, Manger, Fernandez and Rubidge2016b, Reference Benoit2024).

Figure 6. The complex three-dimensional anatomy of the maxillary canal in Choerosaurus (SAM-PK-K 8797) and Pachydectes (BP/1/5735). Skulls in transparent. (1) Lateral and oblique views of the snout and lower jaw of Choerosaurus; (2) lateral, anterior, and oblique views of the snout of Pachydectes.

Accordingly, a role in display has also been hypothesized to account for the diversity of cranial adornments in synapsids, starting with the earliest horned representative of the group, Tetraceratops insignis Matthew, Reference Matthew1908, from the early Permian (Benoit et al., Reference Benoit, Manger, Fernandez and Rubidge2016b). This has been proposed particularly for dinocephalians (Benoit et al., Reference Benoit, Manger, Fernandez and Rubidge2016b, Reference Benoit, Manger, Norton, Fernandez and Rubidge2017e, Reference Benoit2024), given the diversity of their headgear (Fig. 5). Authors who study modern mammals have noted that, if it was only for protecting the brain during head-butting contests, evolving such cranial adaptations would be at odds with the fact that head-on combat causes long-term neurodegenerative issues, but rarely has any immediate consequences on fitness (Ackermans et al., Reference Ackermans, Varghese, Williams, Grimaldi, Selmanovic, Alipour, Balchandani, Reidenberg and Hof2022; Ackermans, Reference Ackermans2023; Woodruff and Ackermans, Reference Woodruff and Ackermans2024; Ackermans and Reidenberg, Reference Ackermans and Reidenberg2025). Because the damage to neurological functions is mostly cumulative, and because head-butting animals only need to survive long enough to reproduce, in terms of pure evolutionary economics, they should not display such exaggeratedly pachyostotic cranial adornments (Ackermans et al., Reference Ackermans, Varghese, Williams, Grimaldi, Selmanovic, Alipour, Balchandani, Reidenberg and Hof2022; Ackermans, Reference Ackermans2023). Why would dinocephalians grow and maintain such enlarged headgears? The answer may lie in the fact that the display of an animal’s headgear is used as a passive show of force during courtship, which places it under sexual selection (Geist, Reference Geist1966; Emlen, Reference Emlen2008).

In modern ungulates, head adornments are used not only for combat, but they are also used to secure mates during ritualized encounters and to attract females during courtship displays (Geist, Reference Geist1966; Woodruff and Ackermans, Reference Woodruff and Ackermans2024). This places head adornments under sexual selection, which often exacerbates traits beyond functional expectations and allows for the evolution of highly expensive structures, such as oversized bosses, horns, and antlers. In comparison, woodpeckers practice intensive head-butting, yet their skulls are far from displaying the degree of pachyostosis seen in head-butting mammals or dinocephalians (Smoliga, Reference Smoliga2024). This illustrates how strengthening the braincase alone is not enough to account for the extreme cranial features observed in head-butting animals and supports that sexual selection also must be at play in these cases (Woodruff and Ackermans, Reference Woodruff and Ackermans2024).

In a manner consistent with the above, the FEA results show that dinocephalian cranial bones may be thicker than required to resist head-butting. They could even be better adapted to head-butting than any modern or extinct animal that has been evaluated through FEA, including pachycephalosaurid dinosaurs (Fig. 7; Snively and Theodor, Reference Snively and Theodor2011; Bolton and Benoit, Reference Benoit2024). The margin of performance between the dinocephalians and controls was so vast that in order to test the effect of absolute skull size, the control models were proportionally enlarged such that their skull lengths would match that of the similarly proportioned Moschops. Even then, only the enlarged Stegoceras model managed to outperform a dinocephalian model (Estemmenosuchus), which is the worst-performing dinocephalian in these analyses (Fig. 7; Bolton and Benoit, Reference Benoit2024).

Figure 7. Force application point stresses of selected skull models from the 3400 N head- and flank-butting protocols, with sagittal sections of Stegoceras and Moschops models (not to scale). Top right: equivalent stress color legend.

It is possible that sexual selection was, thus, involved in the evolution of the grotesquely pachyostosed dinocephalian cranial anatomy, which would support that they practiced sexual display. A function of cranial adornments in thermoregulation has been proposed in modern birds with a casque (Eastick et al., Reference Eastick, Tattersall, Watson, Lesku and Robert2019), but this would imply that dinocephalians were endothermic, which is not the best-supported hypothesis (see below).

The existence of display in dinocephalians would further support the case for life in groups. It would imply ritualized behavior and the ability to interpret an opponent’s intentions and anticipate its moves, which would bolster the claim that synapsids were social animals. Among modern mammals, ritualized agonistic displays (as opposed to combats) are more often practiced in gregarious species because avoiding combats and injuries is vital not to attract predators towards the herd (Geist, Reference Geist1966; Cabrera and Stankowich, Reference Cabrera and Stankowich2020). Accordingly, ungulate species with the largest weapons are the most gregarious ones, whereas those with diminutive but lethally sharp horns and canines are usually more solitary, which, again, compares well with the situation in dinocephalians (Fig. 4).

To conclude, the case for social interactions among many synapsid groups is growing, and now potentially includes evidence for parental care, life in groups, sexual display, and combat. These support the existence of remarkably advanced behavior for these late Paleozoic to early Mesozoic taxa.

Endothermy, hair, milk, and sense organs

A fur pelt is a necessary condition to maintain an elevated body temperature in mammals. Coincidentally, hairs are also associated with mammary glands as the latter evolved from modified apocrine glands located at the base of the former. Moreover, monotremes have no nipple, and milk drops directly from abdominal hairs, where it is sucked by neonates (Oftedal, Reference Oftedal2002, Reference Oftedal, Ogra, Walker and Lönnerdal2020). As such, the evolutionary history of endothermy, hair, and milk production are tightly linked, and since they are the most quintessential mammalian-defining characters, deciphering their origins has attracted vast amounts of research attention. Despite this, little consensus has been achieved concerning how and when they evolved. Because these soft-tissue and physiological traits cannot fossilize, hypotheses about their origins are controversial.

Hair most likely evolved from scales, as suggested by embryology and a handful of pelycosaur fossils preserving scale impressions (Reisz and Reisz, Reference Reisz and Reisz1972; Niedźwiedzki and Bojanowski, Reference Niedźwiedzki and Bojanowski2012; Di-Poï and Milinkovitch, Reference Di-Poï and Milinkovitch2016; Spindler et al., Reference Spindler, Werneburg, Schneider, Luthardt, Annacker and Rößler2018; Marchetti et al., Reference Marchetti, Logghe, Buchwitz and Fröbisch2025). Some rare mummified therapsid (i.e., non-cynodont therapsid) skin fragments support that it was verrucous and scale-less (Chudinov, Reference Chudinov1968; Smith and Botha-Brink, Reference Smith and Botha-Brink2014; Smith et al., Reference Smith, Botha and Viglietti2022; Benoit, Reference Benoit2024). The occurrence of hair is so far limited to crown group Mammalia (Ji et al., Reference Ji, Luo, Yuan and Tabrum2006; Martin et al., Reference Martin, Marugán-Lobón, Vullo, Martín-Abad, Luo and Buscalioni2015; Meng et al., Reference Meng, Grossnickle, Liu, Zhang, Neander, Ji and Luo2017), with no intermediate stages known. Depending on the phylogenetic placement of haramiyidans (within crown group Mammalia or as non-mammalian mammaliaforms), the evidence for a fur pelt may be expanded to early mammaliaforms (Li et al., Reference Li, D’Alba Altamirano, Debruyn, Dobson, Zhou, Clarke, Vinther, Li and Shawkey2025). Filamentous structures preserved in Permian coprolites have been proposed as hair-like, but this was later dismissed by Benoit et al. (Reference Benoit, Manger and Rubidge2016a). The condition in cynodonts, the last forerunners of mammals among Permo-Triassic taxa (Fig. 1), remains so far undocumented, although Ren et al. (Reference Ren, Wang, Wei, Liu, Meng and Mao2025) mentioned possible hair preserved in one Cretaceous tritylodontid from China.

Against the major challenge posed by the poor fossil record of hair, indirect proxies that often loosely correlate with mammalian endothermy have been relied upon in order to trace the evolution of synapsid thermophysiology (Bennett and Ruben, Reference Bennett, Ruben, Hotton, MacLean, Roth and Roth1986; Hillenius and Ruben, Reference Hillenius and Ruben2004). The conflicting results propose opposing timelines, ranging from an origin of endothermy in the last common ancestor of amniotes (e.g., Legendre et al., Reference Legendre, Guénard, Botha-Brink and Cubo2016; Grigg et al., Reference Grigg, Nowack, Bicudo, Bal, Woodward and Seymour2022) to the opposite extreme in which not even early mammaliaforms were endothermic (e.g., Newham et al., Reference Newham, Gill, Brewer, Benton and Fernandez2020, Reference Newham, Gill and Corfe2022). The most consensual hypotheses reconcile these contrasting views by invoking a very slow, incremental evolution of endothermy over long periods of time (Kemp, Reference Kemp2005; Lovegrove, Reference Lovegrove2019; Benton, Reference Benton2021; Faure-Brac et al., Reference Faure-Brac, Woodward, Aubier and Cubo2024), but these remain unsatisfactory because most proxies have been long considered unreliable (Bennett and Ruben, Reference Bennett, Ruben, Hotton, MacLean, Roth and Roth1986). Even the reliability of ossified maxillary turbinates, once considered the “Rosetta Stone of endothermy” (Bennett and Ruben, Reference Bennett, Ruben, Hotton, MacLean, Roth and Roth1986), has recently been questioned (Owerkowicz et al., Reference Owerkowicz, Musinsky, Middleton, Crompton, Brainerd, Dial and Shubin2015; Martinez et al., Reference Martinez, Okrouhlík, Šumbera, Wright, Araújo, Braude, Hildebrandt, Holtze, Ruf and Fabre2023; Fonseca et al., Reference Fonseca, Martinelli, Gill, Rayfield, Schultz, Kerber, Ribeiro, Francischini and Soares2024b).

Given these challenges, new proxies based on paleoneurology were recently introduced to trace the origin of hair and endothermy across synapsid phylogeny: the maxillary canal, parietal foramen, and inner-ear thermo-motility index (Benoit et al., Reference Benoit, Manger, Fernandez and Rubidge2016b, Reference Benoit, Ruf, Miyamae, Fernandez, Rodrigues and Rubidge2020a; Araújo et al., Reference Araújo, David, Benoit, Lungmus and Stoessel2022b; Norton et al., Reference Norton, Abdala and Benoit2023).

The CT-assisted study of the maxillary canal enables tracing the evolution of the infraorbital nerve that innervates the mystacial vibrissae (or facial whiskers) (Benoit et al., Reference Benoit, Manger and Rubidge2016a, Reference Benoit, Ruf, Miyamae, Fernandez, Rodrigues and Rubidge2020a; Miyamae et al., Reference Miyamae, Benoit, Ruf, Sibiya and Bhullar2024). The rami of the maxillary branch of the trigeminal nerve are homologous to the osseous tube within the maxillary bone in modern mammals and reptiles (Benoit et al., Reference Benoit, Manger and Rubidge2016a). In modern reptiles, this tube is a long, narrow, ramified canal that encloses the nerves and blood vessels that innervate and supply the tissues of the upper lip. In most synapsids, the maxillary canal morphology conforms to the reptilian condition, although more branched (Fig. 8). In reptiles, on the one hand, the canal is linear and the supralabial foramina are aligned above the tooth row, whereas the maxillary canal of all synapsids and most cynodonts is branched and complex (Benoit et al., Reference Benoit, Manger and Rubidge2016a, Reference Benoit, Ruf, Miyamae, Fernandez, Rodrigues and Rubidge2020a, Reference Benoit, Ford, Miyamae and Ruf2021b; Miyamae et al., Reference Miyamae, Benoit, Ruf, Sibiya and Bhullar2024; Pusch et al., Reference Pusch, Kammerer and Fröbisch2024). These branches of the maxillary canal can be homologized to the corresponding rami of the maxillary branch of the mammalian trigeminal nerve (Fig. 8), although their exact homology is still not completely agreed upon (Wallace et al., Reference Wallace, Martínez and Rowe2019; Fonseca et al., Reference Fonseca, Martinelli, Gill, Rayfield, Schultz, Kerber, Ribeiro and Soares2024a). In modern mammals, on the other hand, the canal has evolved into the shorter, simpler, and thicker infraorbital foramen, which gives passage to the sensory fibers of the branch of the trigeminal nerve that innervate the whiskers, the infraorbital nerve (Benoit et al., Reference Benoit, Manger and Rubidge2016a, Reference Benoit, Ruf, Miyamae, Fernandez, Rodrigues and Rubidge2020a; Miyamae et al., Reference Miyamae, Benoit, Ruf, Sibiya and Bhullar2024).

Figure 8. The homology of the synapsid maxillary canal to the trigeminal system of mammals. (1) Fragment of a gorgonopsian maxillary bone exposing the natural cast of the maxillary canal, courtesy of the Museum National d’Histoire Naturelle, Paris; (2) the human maxillary branch of the trigeminal nerve. Not to scale.

The reason behind the transition from the maxillary canal to the infraorbital foramen resides in whisking, the ability of mammals to move their facial muscles to use their vibrissae as tactile organs (Benoit et al., Reference Benoit, Manger and Rubidge2016a, Reference Benoit, Ruf, Miyamae, Fernandez, Rodrigues and Rubidge2020a; Miyamae et al., Reference Miyamae, Benoit, Ruf, Sibiya and Bhullar2024). A wide and short infraorbital foramen would let the infraorbital nerve pass through the maxilla and ramify into the soft tissue of the face. In contrast, the vast majority of synapsids could not have whisked because their infraorbital nerve was engulfed entirely by the maxilla, like in modern reptiles (although some variations exist, see e.g., Franco et al., Reference Franco, Müller, Martinelli, Hoffmann and Kerber2021; Benoit et al., Reference Benoit, Araujo, Lund, Bolton, Lafferty, Macungo and Fernandez2024; Pusch et al., Reference Pusch, Kammerer and Fröbisch2024; Medina et al., Reference Medina, Martinelli, Gaetano, Roese-Miron, Tartaglione, Backs, Novas and Kerber2025). Because evolving whiskers would have been pointless without the ability to whisk (Grant et al., Reference Grant, Breakell and Prescott2018), it is safe to hypothesize that synapsids did not have whiskers. The numerous foramina visible on the surface of the maxillary bone in most synapsids (sometimes referred to as “whisker pits”) and that have been mistakenly identified as rooting points for sensory hairs (e.g., Lingham-Soliar, Reference Lingham-Soliar and Lingham-Soliar2014; Bonnan, Reference Bonnan2016) are in fact connected to a reptilian-like maxillary canal. These pits do not indicate the presence of whiskers (Benoit et al., Reference Benoit, Manger and Rubidge2016a, Reference Benoit, Ruf, Miyamae, Fernandez, Rodrigues and Rubidge2020a; Miyamae et al., Reference Miyamae, Benoit, Ruf, Sibiya and Bhullar2024).

Transformation of the maxillary canal into the mammalian infraorbital foramen occurred 241 million years ago, in probainognathian cynodonts (Benoit et al., Reference Benoit, Ruf, Miyamae, Fernandez, Rodrigues and Rubidge2020a). In early probainognathians, such as Lumkuia and Ecteninion, the maxillary canal is very branched (which is the plesiomorphic condition for synapsids), then, in later-diverging probainognathians (i.e., the Prozostrodontia), the maxillary canal gradually became simpler as the infraorbital nerve was freed from the maxillary bone, and by extension, the skull (Fig. 9). Most of the anterior branches of the canal were lost, except the rostral alveolar canal. This simplification of the maxillary canal correlates with the separation of the infraorbital canal from the maxillary sinus and lacrimal canal (although Fonseca et al., Reference Fonseca, Martinelli, Gill, Rayfield, Schultz, Kerber, Ribeiro and Soares2024a, pointed out that some variations still exist in prozostrodontians). The infraorbital canal also became very short compared to the rostral alveolar canal in prozostrodontians, whereas in more basal probainognathians (such as Probainognathus) it is the infraorbital part of the maxillary canal that is the longest (Benoit et al., Reference Benoit, Manger and Rubidge2016a, Reference Benoit, Ruf, Miyamae, Fernandez, Rodrigues and Rubidge2020a; Fonseca et al., Reference Fonseca, Martinelli, Gill, Rayfield, Schultz, Kerber, Ribeiro and Soares2024a; Kerber et al., Reference Kerber, Roese-Miron, Bubadué and Martinelli2024a). This relative dominance of the rostral alveolar canal is still the condition observed in most mammals (Miyamae et al., Reference Miyamae, Benoit, Ruf, Sibiya and Bhullar2024).

Figure 9. Evolution of the maxillary canal (in green) into the mammalian infraorbital foramen across Synapsida. Note the simplification of the branching pattern of the maxillary canal in probainognathians, the appearance of the infraorbital foramen in derived forms, and its enlargement in mammals.

This evolution of the maxillary canal in prozostrodontians indicates that 241 million years ago, their last common ancestor acquired a facial skeletal morphology consistent with the presence of whiskers and could perform whisking (Fig. 9). This is supported by the distribution of facial muscles as evinced by muscle scars on their fossilized skulls, which shows they had acquired facial mobility (Miyamae et al., Reference Miyamae, Benoit, Ruf, Sibiya and Bhullar2024), and by the diameter of their infraorbital foramen (Benoit et al., Reference Benoit, Manger and Rubidge2016a; Muchlinski et al., Reference Muchlinski, Wible, Corfe, Sullivan and Grant2020). A rhinarium evolved with the loss of the internarial process of the premaxilla later in Mammaliaformes, achieving transformation of the facial anatomy (Higashiyama et al., Reference Higashiyama, Koyabu, Hirasawa, Werneburg, Kuratani and Kurihara2021; Benoit et al., Reference Benoit, Dollman, Smith and Manger2023b). The origin of whiskers 241 million years ago supports that the genetic composition to produce hair was present in derived probainognathians some 150 million years prior to the origin of crown group Mammalia (consistent with Ren et al., Reference Ren, Wang, Wei, Liu, Meng and Mao2025).

In parallel, independent evidence for the origin of endothermy came from the study of the viscosity of the inner-ear fluid, the endolymph (Araújo et al., Reference Araújo, David, Benoit, Lungmus and Stoessel2022b). The endolymph flows inside the semicircular canals of the inner ear, which is the balance organ. Endolymph movements inside the semicircular canals, which is oriented in the three dimensions of space, activate sensory receptors that monitor balance (Graf and Klam, Reference Graf and Klam2006; Ekdale, Reference Ekdale2016). As some tetrapod lineages transitioned from a colder body temperature to a warmer one, the viscosity of their endolymph changed, which would have disrupted their balance organ if not checked. In the avian lineage, the balance function was maintained through transition to endothermy by altering the chemical composition of the endolymph (Araújo et al., Reference Araújo, David, Benoit, Lungmus and Stoessel2022b).

In the synapsid lineage, however, fluid viscosity was maintained by reducing the diameter of the cross section of the semicircular canals, thus increasing pressure inside the balance organ. This implies that tracing this change of semicircular canal sectional diameter through synapsid phylogeny would enable us to pinpoint the exact timing of the origin of mammalian endothermy. The dimensions of the semicircular canals of the inner ear are almost the same as those of the bony labyrinth (i.e., the osseous capsules that contain the membranous inner ear within the skull). This means that their cross-sectional diameter can be measured in extinct taxa.

Araújo et al. (Reference Araújo, David, Benoit, Lungmus and Stoessel2022b) designed the thermo-motility index, which removes the locomotory and phylogenetic components from the geometry of the semicircular canals to obtain a metric that is solely proportional to body temperature. Tracing the thermo-motility index through synapsid phylogeny, they found that an almost 10°C increase in body temperature suddenly happened in the probainognathians 233 million years ago (Fig. 10). This is in sharp contrast with the previous models that advocated for a long and slow transition towards endothermy. This date places the origin of endothermy just 8 million years after that of whiskers, again in the probainognathian cynodonts. Interestingly, it coincides with the Carnian Pluvial Episode, a global humid event that caused the extinction of numerous tetrapod lineages and sparked the radiation of dinosaurs (Benton et al., Reference Benton, Bernardi and Kinsella2018; Qvarnström et al., Reference Qvarnström, Vikberg Wernström, Wawrzyniak, Barbacka and Pacyna2024).

Figure 10. Evolution of the thermo-motility index (TMI) in cynodonts though the Permian and Triassic. L.P. = late Permian; Mid. = Middle; M.P. = middle Permian; Mmf. = Mammaliaformes; Prb. = Probainognathia; T. = Triassic. Brackets indicate 95% confidence intervals. Redrawn after Araújo et al. (Reference Araújo, David, Benoit, Lungmus and Stoessel2022b).

A final line of paleoneurological evidence supporting the evolution of hair and endothermy in probainognathian cynodonts is loss of the parietal foramen for the pineal eye (Roth et al., Reference Roth, Roth, Hotton, Hotton, MacLean, Roth and Roth1986; Benoit et al., Reference Benoit, Abdala, Van den Brandt, Manger and Rubidge2015, Reference Benoit, Abdala, Manger and Rubidge2016c). The pineal eye (or third eye) is a photoreceptive organ that, in modern ectotherms, plays a crucial role in making thermoregulatory decisions (Hutchison and Kosh, Reference Hutchison and Kosh1974; Quay, Reference Quay, Gans, Northcutt and Ulinski1979; Ralph et al., Reference Ralph, Firth, Gern and Owens1979; Roth et al., Reference Roth, Roth, Hotton, Hotton, MacLean, Roth and Roth1986). It is located on the skull roof in an opening known as the parietal foramen. Both the parietal foramen and corresponding third eye are absent in mammals, which suggests that their loss may be correlated to the origin of endothermy (Roth et al., Reference Roth, Roth, Hotton, Hotton, MacLean, Roth and Roth1986; Benoit et al., Reference Benoit, Abdala, Van den Brandt, Manger and Rubidge2015).

A systematic survey of the presence or absence of a parietal foramen in synapsids in order to determine the exact timing of the loss of the pineal eye has shown that it is present in most synapsids (Fig. 11). Then, its presence becomes more variable in cynodonts and it finally disappears completely in Lumkuia and other probainognathian cynodonts (Benoit et al., Reference Benoit, Abdala, Manger and Rubidge2016c; Benoit, Reference Benoit2023). Genetic experiments on mice have shown that MSX2, a homeogene that controls the presence of a parietal foramen, also monitors the maintenance of a fur pelt (Satokata et al., Reference Satokata, Ma, Ohshima, Bei and Woo2000; Garcia-Miñaur et al., Reference Garcia-Miñaur, Mavrogiannis, Rannan-Eliya, Hendry, Liston, Porteous and Wilkie2003; Agarwal et al., Reference Agarwal, Pandey, Baranwal and Roy2015). When the gene is mutated, mice with this mutation do not have hair, and their skull is perforated by a parietal foramen. Coupled with the thermoregulatory role played by the third eye in reptiles, these experiments support that the loss of the parietal foramen may be linked to the origin of fur (Benoit et al., Reference Benoit, Manger and Rubidge2016a, Reference Benoit, Abdala, Manger and Rubidgec). Pleiotropically, the same mutation of MSX2 that causes mice to acquire a parietal foramen and fail to maintain their fur coverage also makes them lose their mammary glands (Satokata et al., Reference Satokata, Ma, Ohshima, Bei and Woo2000), making them essentially “de-evolve” into a synapsid-like physiology. This is not surprising given the multiple evolutionary connections between hair and mammary glands mentioned above (Oftedal, Reference Oftedal2002, Reference Oftedal, Ogra, Walker and Lönnerdal2020).

Figure 11. Evolution of the presence of the parietal foramen in synapsids and associated changes in paleobiology. The frequency (Fq) of the presence of a parietal foramen is given for each clade in %. The number (n) of specimens studied is indicated between brackets. IOF = infraorbital foramen; Jur. = Jurassic.

Furthermore, this mutation of MSX2 has another side effect on the cerebellum. In mice, when MSX2 is mutated, the vermis of the cerebellum shrinks, whereas in wild-type mice, the vermis is well developed. This is consistent with the condition in cynodonts: in early cynodonts, there is no trace of the cerebellar vermis on the endocast, like in mutated mice, but in Lumkuia (the basal-most probainognathian and the first to lose its parietal foramen; Benoit et al., Reference Benoit, Nxumalo, Norton, Fernandez, Gaetano, Rubidge and Abdala2022), the expanded cerebellar vermis is salient on the dorsal aspect of the endocast (Fig. 12), like in wild-type mice (Benoit et al., Reference Benoit, Manger and Rubidge2016a; Benoit, Reference Benoit2023). Overall, this is yet more independent evidence that supports that a crucial mutation of MSX2 in probainognathians was responsible for the loss of the parietal foramen, development of the cerebellum, and acquisition of hair and mammary glands, which rapidly enhanced their “mammalness” in the Triassic (Fig. 11).

Figure 12. Evolution of the endocast in cynodonts, in oblique view, skulls in transparent. (1) The basal cynodont Procynosuchus; (2) the early probainognathian Lumkuia; (3) the early mammaliaform Megazostrodon. Not to scale.

Altogether, the CT-assisted study of the maxillary canal, endolymph viscosity, and evolutionary development of the parietal foramen and cerebellar vermis led to the emergence of a new, very consistent evolutionary hypothesis for the origin of hair, endothermy, and mammary glands. In this hypothesis, significant mutation of MSX2 occurred in the probainognathian lineage, as demonstrated by the loss of their parietal foramen and enlarged cerebellar vermis (Benoit et al., Reference Benoit, Nxumalo, Norton, Fernandez, Gaetano, Rubidge and Abdala2022, Reference Benoit, Dollman, Smith and Manger2023b), which turned the probainognathians into more mammal-like animals about 247 million years ago (early Anisian). The appearance of an interhemispheric sulcus between the two cerebral hemispheres in Lumkuia (Fig. 12) is consistent with an enlargement of the somatosensory cortex due to the establishment of a fur pelt (Rowe, Reference Rowe and Kaas2020, Reference Rowe, Dozo, Paulina-Carabajal, Macrini and Walsh2023; Benoit, Reference Benoit2023). This event would have been followed rapidly by the appearance of whiskers 241 million years ago, and a little later by the onset of endothermy 233 million years ago (as indicated by the thermo-motility index; Araújo et al., Reference Araújo, David, Benoit, Lungmus and Stoessel2022b). As a consequence, the appearance of hair likely predated the origin of endothermy, and because MSX2 is involved in maintaining the fur pelt only (not in its development), it is possible that hair was present before the origin of probainognathians, although evidence is lacking (Benoit et al., Reference Benoit, Manger and Rubidge2016a).

Our hypothesis posits that mammary glands evolved early in probainognathian history, as indicated by MSX2; however, some of the most derived probainognathians, the tritylodontids, had erupted teeth at birth, thus were likely self-sufficient enough not to rely on maternal milk (Hoffman and Rowe, Reference Hoffman and Rowe2018; Benoit, Reference Benoit2019). This would imply that the appearance of mammary glands predated milk sucking, the latter of which likely evolved in early mammaliaforms coincidentally with diphyodonty (Norton et al., Reference Norton, Abdala and Benoit2023). All previous authors agree that lactating did not initially evolve for feeding neonates but as skin secretions used, for example, for hormonal signaling or to moisturize the eggs (see Oftedal, Reference Oftedal2002, Reference Oftedal, Ogra, Walker and Lönnerdal2020, for reviews). Our hypothesis is consistent with this notion; however, it is noteworthy that milk-feeding in tritylodontids is not completely excluded as some rodent species have erupted teeth at birth that already show signs of wear (Sone et al., Reference Sone, Koyasu, Kobayashi, Kobayashi and Oda2008).

The synchrotron study of dentine increments in early mammaliaforms, in contrast, recently raised questions regarding the above evolutionary hypothesis by unravelling the surprising longevity and slow growth rates of these shrew-sized animals. These cast doubts on whether mammaliaforms such as Morganucodon were truly endothermic (Newham et al., Reference Newham, Gill, Brewer, Benton and Fernandez2020, Reference Newham, Gill and Corfe2022). More recent data, however, have shown that this condition was generalized to most Mesozoic mammaliaforms, including representatives of crown group Mammalia (Newham et al., Reference Newham, Corfe, Brewer, Bright and Fernandez2024; Panciroli et al., Reference Panciroli, Benson, Fernandez, Fraser, Humpage, Luo, Newham and Walsh2024).

In parallel, it was proposed that Mesozoic mammals went through a longevity bottleneck (de Magalhães, Reference de Magalhães2023) during which the miniaturization of body size and increased metabolic rate did not affect longevity. Mesozoic mammals would have been comparable to songbirds (Passeriformes), which are long-lived, yet small-bodied and endothermic (Wasser and Sherman, Reference Wasser and Sherman2010). In songbirds, this is due to efficient predator-avoidance strategies (i.e., flight) that make them survive long enough to preserve the selective advantage of maintaining their genes against senescence (Shattuck and Williams, Reference Shattuck and Williams2010; Wasser and Sherman, Reference Wasser and Sherman2010). Mesozoic mammals did not benefit from this protection, and it follows that their longevity would have eventually decreased due to extreme levels of predation (by the dominant archosaurs), which dropped their life expectancy and thus cancelled the selective pressure against senescence (de Magalhães, Reference de Magalhães2023). The longevity bottleneck hypothesis would thus elegantly reconcile the opposing signals obtained from the thermo-motility index and dentine by supporting that early mammaliaforms were indeed endothermic, but not yet short-lived. Even more new proxies for endothermy are being developed each year (Legendre and Davesne, Reference Legendre and Davesne2020; Wiemann et al., Reference Wiemann, Menéndez, Crawford, Fabbri, Gauthier, Hull, Norell and Briggs2022) and will address whether this hypothesis withstands the test of time.

Synapsid endocasts and encephalization quotient

Mammals have the most folded and largest brains (in absolute and relative size) in the animal kingdom (Manger et al., Reference Manger, Spocter and Patzke2013). This condition is derived from the synapsid one, in which the endocast is small, smooth, linearly arranged, and tubular (Fig. 12). A lot of the paleoneurological literature has been dedicated to studying the transition from the primitive reptilian-like endocast of early synapsids to the mammalian condition (Boonstra, Reference Boonstra1968; Jerison, Reference Jerison1973; Edinger, Reference Edinger1975; Hopson et al., Reference Hopson, Gans, Northcutt, Ulinski, Gans, Northcutt and Ulinski1979; Quiroga, Reference Quiroga1979, Reference Quiroga1980, Reference Quiroga1984; Kielan-Jaworowska, Reference Kielan-Jaworowska1983, Reference Kielan-Jaworowska1984; Durand, Reference Durand1989; Kielan-Jaworowska et al., Reference Kielan-Jaworowska, Cifelli and Luo2004; Rowe et al., Reference Rowe, Macrini and Luo2011; Castanhinha et al., Reference Castanhinha, Araújo, Júnior, Angielczyk, Martins, Martins, Chaouiya, Beckmann and Wilde2013; Ruf et al., Reference Ruf, Maier, Rodrigues and Schultz2014; Laaß, Reference Laaß2015a; Benoit et al., Reference Benoit, Fernandez, Manger and Rubidge2017b, Reference Benoit, Dollman, Smith and Manger2023b; Rodrigues et al., Reference Rodrigues, Martinelli, Schultz, Corfe, Gill, Soares and Rayfield2018; Pavanatto et al., Reference Pavanatto, Kerber and Dias‐da‐Silva2019; de Simão-Oliveira et al., Reference de Simão-Oliveira, Kerber and L. Pinheiro2020; Rowe, Reference Rowe and Kaas2020, Reference Rowe, Dozo, Paulina-Carabajal, Macrini and Walsh2023; Hoffmann et al., Reference Hoffmann, Rodrigues, Soares and Andrade2021; Kerber et al., Reference Kerber, Ferreira, Fonseca, Franco, Martinelli, Soares and Ribeiro2021, Reference Kerber, Roese-Miron, Bubadué and Martinelli2024a; Benoit, Reference Benoit2023; de Sousa et al., Reference de Sousa, Beaudet, Calvey, Bardo and Benoit2023; Pusch et al., Reference Pusch, Kammerer and Fröbisch2024; Medina et al., Reference Medina, Martinelli, Gaetano, Roese-Miron, Tartaglione, Backs, Novas and Kerber2025). Fueled by CT data, the study of synapsid, mammaliaform, and mammalian brain size and encephalization in the past 10 years has grown faster than ever (Rowe et al., Reference Rowe, Carlson, Bottorff and Olson1995, Reference Rowe, Macrini and Luo2011; Rodrigues, Reference Rodrigues2005; Macrini, Reference Macrini2006; Macrini et al., Reference Macrini, Rougier and Rowe2007; Du Plessis, Reference Du Plessis2010; Hoffmann et al., Reference Hoffmann, O’Connor, Kirk, Wible and Krause2014; Rodrigues et al., Reference Rodrigues, Ruf and Schultz2014, Reference Rodrigues, Martinelli, Schultz, Corfe, Gill, Soares and Rayfield2018; Laaß, Reference Laaß2015a; Araújo et al., Reference Araújo, Fernandez, Polcyn, Fröbisch and Martins2017; Benoit et al., Reference Benoit, Fernandez, Manger and Rubidge2017b, Reference Benoit, Dollman, Smith and Manger2023b; Laaß and Kaestner, Reference Laaß and Kaestner2017; Csiki-Sava et al., Reference Csiki-Sava, Vremir, Meng, Brusatte and Norell2018; Wallace, Reference Wallace2018; Pavanatto et al., Reference Pavanatto, Kerber and Dias‐da‐Silva2019; Hoffmann et al., Reference Hoffmann, Rodrigues, Soares and Andrade2021; Kerber et al., Reference Kerber, Ferreira, Fonseca, Franco, Martinelli, Soares and Ribeiro2021, Reference Kerber, Roese-Miron, Bubadué and Martinelli2024a, Reference Kerber, Roese‐Miron, Medina, da Roberto‐da‐Silva, Cabreira and Prettob, Reference Kerber, Müller, De Simão‐Oliveira, Pretto and Martinelli2025; Pusch et al., Reference Pusch, Kammerer, Fernandez and Fröbisch2022, Reference Pusch, Kammerer and Fröbisch2024; Gigliotti et al., Reference Gigliotti, Pusch, Kammerer, Benoit and Fröbisch2023; Macungo et al., Reference Macungo, Araújo, Browning, Smith, David, Angielczyk, Massingue, Ferreira-Cardoso, Kortje and Lee2023; George et al., Reference George, Kammerer, Foffa, Clark and Brusatte2024; Benoit, Reference Benoit2025; Benoit and Jodder, Reference Benoit and Jodder2025; Medina et al., Reference Medina, Martinelli, Gaetano, Roese-Miron, Tartaglione, Backs, Novas and Kerber2025; Ren et al., Reference Ren, Wang, Wei, Liu, Meng and Mao2025). Since the seminal work by Jerison, the dataset of measured endocranial volumes has grown from 3 documented synapsid genera in 1973, to 7 in 2011 (Rowe et al., Reference Rowe, Macrini and Luo2011), 24 in 2017 (Benoit et al., Reference Benoit, Fernandez, Manger and Rubidge2017b), to over 50 today, including 32 cynodont genera (Table 2).

Table 2. Body Mass (BM), Endocast Volume including olfactory bulbs (EV), Synapsid Encephalization Quotient (SEQ) and ratio of the olfactory bulbs volume over that of the endocast (OB/EV) per genera (data from Table 1). Trirachodon: note that EV without OB was used for this taxon

An encephalization quotient (EQ) is a mathematical tool created to compare brain sizes in tetrapods of vastly different body mass (Jerison, Reference Jerison1973). It is a ratio of the observed brain mass (or endocast volume) of an animal over its predicted brain mass (or endocast volume) given its body mass. The latter is based on a regression line of brain mass (or endocast volume) over body mass of closely related taxa. An EQ of 1 indicates that an animal of a given brain and body mass has the expected brain size for an average animal of this body mass. An EQ above or below 1 means that the animal has a larger or smaller brain than expected for its body mass, respectively. Many EQ formulas have been designed since Jerison’s pioneering work on tetrapod brain evolution, but none of them was adapted to compare mammals (which usually have a large brain) to non-mammalian synapsids (which usually have a much smaller brain, more like that of modern reptiles). The new, large dataset of synapsid endocast volumes prompted the creation of the Synapsid Encephalization Quotient (SEQ) by Benoit et al. (Reference Benoit, Dollman, Smith and Manger2023b) to enable direct comparisons between mammalian and synapsid endocranial size variations. The SEQ equation is:

$$ \mathrm{SEQ}=\mathrm{Endocast}\ \mathrm{Volume}\ \mathrm{in}\;{\mathrm{cm}}^3/{10}^{\left(0.6669\times \mathrm{Log}\;\mathrm{Body}\ \mathrm{Mass}\ \mathrm{in}\;\mathrm{g}\right)\hbox{-} 1.8188} $$

Many methods have been employed by previous authors to estimate endocranial volumes and body weights in synapsids (see references cited in Table 1). In recent years, authors’ approach to studying endocast evolution have become increasingly standardized as CT scanning is more consistently used to measure endocranial volumes, and the equations to estimate body masses are being refined (Castanhinha et al., Reference Castanhinha, Araújo, Júnior, Angielczyk, Martins, Martins, Chaouiya, Beckmann and Wilde2013; Benoit et al., Reference Benoit, Fernandez, Manger and Rubidge2017b, Reference Benoit, Dollman, Smith and Manger2023b; Pavanatto et al., Reference Pavanatto, Kerber and Dias‐da‐Silva2019; Kerber et al., Reference Kerber, Ferreira, Fonseca, Franco, Martinelli, Soares and Ribeiro2021, Reference Kerber, Roese-Miron, Bubadué and Martinelli2024a, Reference Kerber, Roese‐Miron, Medina, da Roberto‐da‐Silva, Cabreira and Prettob, Reference Kerber, Müller, De Simão‐Oliveira, Pretto and Martinelli2025; Pusch et al., Reference Pusch, Kammerer, Fernandez and Fröbisch2022, Reference Pusch, Kammerer and Fröbisch2024; Gigliotti et al., Reference Gigliotti, Pusch, Kammerer, Benoit and Fröbisch2023; Medina et al., Reference Medina, Martinelli, Gaetano, Roese-Miron, Tartaglione, Backs, Novas and Kerber2025). Regarding data from the literature, the standard deviation of the calculated SEQ in the most documented genera ranges from about 0.1 (in Andescynodon and Probainognathus) to 0.4 (in Massetognathus, Thrinaxodon, and Exaeretodon). In Probainognathus, for which multiple methods were used to obtain the body mass and endocranial volume from a single specimen (Table 1), the standard deviation remains low (0.12), which supports that the variations observed are not too skewed by methodological bias. It is not possible to determine how much of the endocranial space was occupied by the actual brain in Permo-Triassic synapsids (Benoit and Midzuk, Reference Benoit and Midzuk2024). Some authors have proposed methods to estimate the percentage of endocranial space occupied by non-neural tissues, in some cases based on the imprints left by the olfactory bulbs, cerebral hemispheres, pineal body, and cerebellum on the endocast (e.g., Quiroga, Reference Quiroga1980, Reference Quiroga1984; Benoit et al., Reference Benoit, Fernandez, Manger and Rubidge2017b). The arbitrary nature of these estimates, however, makes them unreliable and they have been abandoned in most recent studies (Pavanatto et al., Reference Pavanatto, Kerber and Dias‐da‐Silva2019; Kerber et al., Reference Kerber, Ferreira, Fonseca, Franco, Martinelli, Soares and Ribeiro2021, Reference Kerber, Roese-Miron, Bubadué and Martinelli2024a, Reference Kerber, Roese‐Miron, Medina, da Roberto‐da‐Silva, Cabreira and Prettob, Reference Kerber, Müller, De Simão‐Oliveira, Pretto and Martinelli2025; Pusch et al., Reference Pusch, Kammerer, Fernandez and Fröbisch2022, Reference Pusch, Kammerer and Fröbisch2024; Benoit et al., Reference Benoit, Dollman, Smith and Manger2023b; Gigliotti et al., Reference Gigliotti, Pusch, Kammerer, Benoit and Fröbisch2023; Medina et al., Reference Medina, Martinelli, Gaetano, Roese-Miron, Tartaglione, Backs, Novas and Kerber2025).

Unlike previous results (Rowe et al., Reference Rowe, Macrini and Luo2011; Rowe and Shepherd, Reference Rowe and Shepherd2016; Huttenlocker et al., Reference Huttenlocker, Grossnickle, Kirkland, Schultz and Luo2018), the most recent studies that used the SEQ did not find evidence for a three-pulse increase in early mammaliaforms encephalization driven by olfaction (Benoit, Reference Benoit2023; Benoit et al., Reference Benoit, Dollman, Smith and Manger2023b, Reference Benoit2024; Gigliotti et al., Reference Gigliotti, Pusch, Kammerer, Benoit and Fröbisch2023; Fonseca et al., Reference Fonseca, Martinelli, Gill, Rayfield, Schultz, Kerber, Ribeiro and Soares2024a; Kerber et al., Reference Kerber, Roese-Miron, Bubadué and Martinelli2024a; Medina et al., Reference Medina, Martinelli, Gaetano, Roese-Miron, Tartaglione, Backs, Novas and Kerber2025; Ren et al., Reference Ren, Wang, Wei, Liu, Meng and Mao2025). Instead, SEQ values vary mostly randomly around 1.0 in cynodonts (Fig. 13), except in crown group Mammalia in which it increases and stabilizes above 1.5. Olfactory bulb size, in contrast, increases more stemwards in the probainognathians, which retain a SEQ below 1.5 but develop significantly larger olfactory bulb volume compared to other cynodonts (Kolmogorov–Smirnov test: P = 0.004; Fig. 14). This is exemplified by low SEQ genera such as Brasilodon (syn. Brasilitherium) and Lumkuia, which display oversized olfactory bulbs (Fig.s 12, 14; Rodrigues et al., Reference Rodrigues, Ruf and Schultz2014; Benoit, Reference Benoit2023).

Figure 13. The Synapsid Encephalization Quotient (SEQ) and ratio of olfactory bulbs volume over endocranial volume (OB/EV) across synapsids (data from Table 1, and phylogeny as in Fig. 2). Colors: therapsids are in orange, cynodonts in pink, mammaliaforms and mammals in shades of blue, and the OB/EV ratio is in dark gray.

Figure 14. Box-plot of variations of olfactory bulbs volume over endocranial volume (OB/EV) in cynodonts. Points are outliers; brackets indicate 95% confidence intervals.

The “outside-in” model of the origin of the mammalian brain (Fig. 15) was proposed to replace the previously established “three pulses” model (Benoit, Reference Benoit2023; Benoit et al., Reference Benoit, Dollman, Smith and Manger2023b). Firstly, basal cynodonts have a tubular and, on average, rather small brain, although isolated taxa such as Galesaurus, Cistecynodon, and Siriusgnathus could evolve a larger brain (Fig. 13). In the Triassic, terrestrial ecosystems became increasingly dominated by the rising archosaurs, and the climate changed from relatively cool and arid to hot and humid during the Carnian Pluvial Episode, 233 million years ago. The climate then rapidly reversed back into cool and arid, thus severely affecting terrestrial ecosystems (Benton et al., Reference Benton, Bernardi and Kinsella2018; Qvarnström et al., Reference Qvarnström, Vikberg Wernström, Wawrzyniak, Barbacka and Pacyna2024). The rather rare and poorly diversified probainognathian cynodonts faced these changing environmental conditions by becoming endothermic, nocturnal, and smaller (miniaturization), likely as a means to avoid predators and competitors (Jerison, Reference Jerison1973; McNab, Reference McNab1978; Gerkema et al., Reference Gerkema, Davies, Foster, Menaker and Hut2013; Angielczyk and Schmitz, Reference Angielczyk and Schmitz2014; Lautenschlager et al., Reference Lautenschlager, Gill, Luo, Fagan and Rayfield2018; Benton, Reference Benton2021; Araújo et al., Reference Araújo, David, Benoit, Lungmus and Stoessel2022b; Kaiuca et al., Reference Kaiuca, Martinelli, Schultz, Fonseca, Tavares and Soares2024; Li et al., Reference Li, D’Alba Altamirano, Debruyn, Dobson, Zhou, Clarke, Vinther, Li and Shawkey2025).

Figure 15. The “outside-in” model illustrated on the transparent skulls of the basal cynodont Galesaurus (1), early probainognathian Lumkuia (2), and early mammaliaform Megazostrodon (3). Bottom row shows magnified inner ears of the same taxa in lateral view. Not to scale.

These changes happened 240–210 million years ago (see above) and dramatically altered the sensory abilities of the probainognathians. The importance of sight gradually decreased as probainognathians adapted to a nocturnal, low-light environment. As a consequence, they lost their parietal foramen for the pineal eye and sclerotic ring (Angielczyk and Schmitz, Reference Angielczyk and Schmitz2014; Benoit et al., Reference Benoit, Abdala, Manger and Rubidge2016c). This loss of visual sensory inputs was compensated by the development of smell, facial touch, and hearing in probainognathians (Jerison, Reference Jerison1973; Benoit et al., Reference Benoit, Manger and Rubidge2016a, Reference Benoit, Ruf, Miyamae, Fernandez, Rodrigues and Rubidge2020a, Reference Benoit, Dollman, Smith and Manger2023b; Benoit, Reference Benoit2023). This marked the second phase of the “outside-in” model, which was nothing short of a neurosensory revolution (Fig. 15). Starting with Lumkuia (Benoit, Reference Benoit2023), the olfactory bulbs became consistently larger across Probainognathia (Fig. 14) and the maxillary canal began transforming into the infraorbital foramen for the whiskers (Fig. 9). In the ear region, the cochlear recess began inflating and lengthening to become a proper cochlea (Fig. 15; Luo, Reference Luo2001; Benoit et al., Reference Benoit, Dollman, Smith and Manger2023b). The evolution of the cochlea was further fueled by hearing capabilities being skewed towards higher frequencies (Luo et al., Reference Luo, Schultz, Ekdale, Clack, Fay and Popper2016; Schultz, Reference Schultz2025). Early synapsids were generally adapted to hearing low-frequency airborne sounds and ground vibrations (Barry, Reference Barry1968; Allin and Hopson, Reference Allin, Hopson, Webster, Popper and Fay1992; Laaß, Reference Laaß2015c; Luo et al., Reference Luo, Schultz, Ekdale, Clack, Fay and Popper2016; Benoit et al., Reference Benoit, Manger, Fernandez and Rubidge2017c; Bazzana et al., Reference Bazzana, Evans, Bevitt and Reisz2022), but the shrinking body size, smaller middle-ear bones, and shorter interaural distance in derived probainognathians and early mammaliaforms imposed adaptation to higher frequency hearing (Heffner and Heffner, Reference Heffner, Heffner, Webster, Popper and Fay1992; Rodrigues et al., Reference Rodrigues, Ruf and Schultz2014; Luo et al., Reference Luo, Schultz, Ekdale, Clack, Fay and Popper2016; Heffner and Heffner, Reference Heffner and Heffner2018; Rawson et al., Reference Rawson, Martinelli, Gill, Soares, Schultz and Rayfield2024). In parallel, the semicircular canals became smaller to accommodate the more viscous endolymphatic fluids due to the evolution of endothermy (Araújo et al., Reference Araújo, David, Benoit, Lungmus and Stoessel2022b).

Despite these sensory innovations, at this phylogenetic stage, brain size did not change significantly yet, as shown by the SEQ, which remains under 1.5 in probainognathians (Fig. 13). The presence of a shallow interhemispheric sulcus is noted in Lumkuia (Fig. 12) and more derived probainognathians, which suggests the hemispheres and cerebellum are only beginning to grow larger and filling up the endocranial space more completely (Rowe, Reference Rowe and Kaas2020, Reference Rowe, Dozo, Paulina-Carabajal, Macrini and Walsh2023; Benoit, Reference Benoit2023; Benoit et al., Reference Benoit, Dollman, Smith and Manger2023b).

The last phase of the “outside-in” model happened when the tidal wave of changes reached the brain itself, in Mammaliaformes. After a 40-million-year-long period of sense organs remodeling to better process the new sensory inputs discussed above, the cerebral hemispheres became globular and grew so that the SEQ reached 1.5 and stabilized above this value in the Jurassic (Fig.s 12, 13, 15). This was the final step of this sensory revolution brought about by the transition from a diurnal to a nocturnal lifestyle, miniaturization, endothermy, and a complete environmental and ecological shift.

This model differs from the “three pulses” one (Rowe et al., Reference Rowe, Macrini and Luo2011) because it advocates that all the sensory organs played a role in the increase of encephalization (not olfaction only) and shifts the taxonomic focus stemwards in the phylogenetic tree towards the Probainognathia rather than the Mammaliaformes. It offers a new working hypothesis for the origin of the enlarged mammalian brain, in accordance with the latest data. The field of cynodont paleoneurology is moving fast, with new endocasts being described yearly (Medina et al., Reference Medina, Martinelli, Gaetano, Roese-Miron, Tartaglione, Backs, Novas and Kerber2025; Ren et al., Reference Ren, Wang, Wei, Liu, Meng and Mao2025) and new EQs being proposed (see, e.g., the phylogenetic EQ of Kerber et al., Reference Kerber, Roese-Miron, Bubadué and Martinelli2024a). These developments will challenge this model and address its explanatory power.

Future prospects on neurosensory outliers

Over the past 150 years, it has become clear that many defining mammalian characters evolved well before the origin of crown group Mammalia, including endothermy, hair, whiskers, lactation, the isocortex, and enlarged brain and olfactory bulbs. This gives the synapsids, and more particularly the probainognathian cynodonts, a pivotal role in the origin of “mammalness” (Fig. 11). While the picture of the origin of mammalian-defining traits is becoming more and more precise, it is also becoming clearer that more scientific focus on the paleoneurology and behavior of the probainognathians will be needed to fully grasp the complexity of the transition from an essentially reptilian-like biology to a mammalian one. For example, as highlighted above, the questions of whether cynodonts were laying and incubating eggs, remain open. Fossils preserving contact-incubation behavior would, critically, provide evidence for endothermy, like in dinosaurs (Hogan and Varricchio, Reference Hogan and Varricchio2023; Hogan, Reference Hogan2024).

Synapsids dominated terrestrial ecosystems between the late Carboniferous and Early Triassic, and, accordingly, their neurosensory adaptations reflected the diversity of their adaptations (Benoit et al., Reference Benoit, Fernandez, Manger and Rubidge2017b, Reference Benoit2024). Synapsids are not only the ancestors of mammals, but they represent an evolutionary radiation of early amniotes in their own right. The classical picture of mammalian evolution depicts synapsids as mere links in a chain of transformation ultimately leading to mammals, but this fails to capture their diversity. This is well exemplified by the evolution of the synapsid postdentary bones into the middle-ear ossicles, which despite being a textbook example of macroevolution that has been known and taught for almost a century (Luo, Reference Luo2011; Luo et al., Reference Luo, Schultz, Ekdale, Clack, Fay and Popper2016; Maier and Ruf, Reference Maier and Ruf2016), recently proved its limitations by failing to account for numerous instances of convergent evolution and reversals in both cynodonts and early mammals (Meng et al., Reference Meng, Bi, Zheng and Wang2018; Rawson et al., Reference Rawson, Martinelli, Gill, Soares, Schultz and Rayfield2024).

Extreme cases of neurosensory adaptations illustrate the immense diversity of sense organs in synapsids as exemplified by the highly branched maxillary canal in Choerosaurus and Pachydectes (Fig. 6) and strongly re-oriented braincase in tapinocephalid dinocephalians (Fig. 4.1). Synapsids experimented with original characters and adaptations throughout their geographical and ecological range across millions of years. Obligate fossoriality is another example of such adaptations. It has long been thought to be unique to the cistecephalids until similar adaptations were documented in the cynodont Cistecynodon parvus Brink and Kitching, Reference Brink and Kitching1953, from the Early Triassic (Benoit et al., Reference Benoit2024; Lund, Reference Lund2024). This cynodont shows the same enlarged stapedial footplate surface area, inflated vestibule, reduced parietal foramen, and enlarged endocast as Kawingasaurus, which supports that it may have been an obligate fossorial cynodont (Fig. 16). The Late Triassic Boreogomphodon and Middle Jurassic Bienotheroides show similar adaptations (Pusch et al., Reference Pusch, Kammerer and Fröbisch2024; Ren et al., Reference Ren, Wang, Wei, Liu, Meng and Mao2025), thus supporting that obligate fossoriality may not have been rare in cynodonts.

Figure 16. Neurosensory adaptation to obligate fossoriality in synapsids. Brain endocast in pink, inner ear in green, and skull in transparent. (1) Basal, non-fossorial cynodont Procynosuchus; (2) basal, obligate fossorial cynodont Cistecynodon; (3) cistecephalid dicynodont Kawingasaurus.

Understudied groups of synapsids, such as the biarmosuchians and therocephalians, display some of the most specialized neurosensory organs observed so far (e.g., Choerosaurus and Pachydectes are therocephalians and biarmosuchians, respectively; Fig. 6). Biarmosuchians have a unique balance organ (Benoit et al., Reference Benoit, Manger, Fernandez and Rubidge2017c) in which a secondary common crus (a well-established symplesiomorphy of synapsids and mammals; Ekdale, Reference Ekdale2013, Reference Ekdale2016) is absent. Euchambersia mirabilis Broom, Reference Broom1931, a therocephalian, is well known for its maxillary fossae (Benoit, Reference Benoit2016), which have been hypothesized to contain a venomous gland; however, it is possible, given that these fossae are connected to the maxillary canal for the trigeminal nerve, that they may have accommodated a sensory organ instead (Benoit et al., Reference Benoit, Norton, Manger and Rubidge2017d), or a gland for secreting hormones involved in some unknown social behavior (Benoit, Reference Benoit2016). Euchambersia, Choerosaurus, and the other therocephalians with unusual cranial structures (e.g., Tatarinov, Reference Tatarinov1965; Huttenlocker et al., Reference Huttenlocker, Sidor and Angielczyk2015; Liu and Abdala, Reference Liu and Abdala2022) provide a glimpse into the yet untapped depth of their neurosensory diversity. These examples showcase the potential paleoneurological progress that remains to be made in synapsids, which may feed new, promising avenues of research as we explore the phylogenetic tree away from mammalian origins. More intense work in these directions may lead to surprising results.

Finally, dicynodonts also have a uniquely complex and branched maxillary canal, expanding anteriorly within the premaxillary bone because of the presence of a cornified beak (Benoit et al., Reference Benoit, Angielczyk, Miyamae, Manger, Fernandez and Rubidge2018). In modern mammals, the premaxillary bone is embryologically derived from the maxillary prominence of the mandibular arch because it is innervated by the maxillary branch of the trigeminal nerve (Higashiyama et al., Reference Higashiyama, Koyabu, Hirasawa, Werneburg, Kuratani and Kurihara2021). In other vertebrates, the premaxilla is innervated by the ophthalmic branch of the trigeminal nerve and, thus, is likely not homologous to the mammalian premaxilla (Higashiyama et al., Reference Higashiyama, Koyabu, Hirasawa, Werneburg, Kuratani and Kurihara2021). The anterior expansion of the maxillary canal into the premaxilla is not unique to dicynodonts as it is found in specimen BP/1/7199 of the basal cynodont Thrinaxodon (Benoit et al., Reference Benoit, Angielczyk, Miyamae, Manger, Fernandez and Rubidge2018), which suggests that the premaxilla in at least some therapsids had already been replaced by the maxillary prominence of the mandibular arch, as in modern mammals. This will have to be investigated further as it questions the homology of the premaxilla in amniotes and demonstrates the relevance of synapsid paleoneurology for addressing questions beyond the field of paleontology.