The presence of the hyomandibular pouch in other “ostracoderms” still remains controversial (Stensiö, 1927, 1964; Halstead, 1971; Janvier, 1996b,2004; Mallatt, 1996). In heterostracans, the foremost gill compartment impression is always situated behind the eye, and is classically interpreted as the hyomandibular pouch (Figure 4E) (Stensiö, 1964; Novitskaya, 1983, 2015; Mallatt, 1996), though only on the basis of a topological argument (Janvier, 2004). By contrast, Halstead (Halstead, 1971, 1973) thought that a true spiracle has been present in some amphiaspid heterostracans, a group of cyathaspidiforms from the Lower Devonian of Siberia, which possess a pair of bean-shaped openings in the carapace immediately lateral to the eyes (Figures 4G,H), a position moreover exactly as in the living benthonic elasmobranchs. However, most authors think it unlikely that these paired openings are equivalent to the spiracles of gnathostomes (Burrow et al., 2020). They were later interpreted as either displaced openings of the prenasal sina (Figure 4F; Janvier, 1974), or adorbital openings (Figure 4G; Janvier, 1996b), probably with an inhalent function for intake of clear water since these amphiaspids were benthic animals (Janvier, 1996b). As the neurocranial anatomy of heterostracans cannot be reconstructed in detail, they will not be considered further here.

In osteostracans, the entire branchial apparatus uniquely shifted forward so that the first two branchial gill pouches lie anterior to the eyes (Figures 4B,C; Stensiö, 1927, 1964; Jarvik, 1980; Kuratani and Ahlberg, 2018). Therefore, it is difficult to determine the hierarchy of the branchial arches according to their topological positions. Mallatt (1996) considered that the foremost gill compartment was the hyomandibular pouch, whereas Janvier (Janvier, 1985, 1996b, 2007) thought that there is no reason to assume that they possessed a prespiracular or even a spiracular gill-pouch (Figures 4B,C) since the mandibular arch bore a velum as in lampreys. In osteostracans, the canal for the facial nerve fused with the trigeminal chamber in the postero-ventral part of orbital cavity (Janvier, 1981; trig, Figures 13–17). Thus, it still cannot be determined whether the mandibular branch of trigeminal nerve passed through the foremost canal, together with the maxillary branch (Figure 4D), or passed through the second canal, together with the facial nerve (Figure 4B). Janvier preferred the latter interpretation, thus the mandibular arch and the hyoid arch of osteostracans are probably closely associated or perhaps fused into a single arch (Figures 4B,C), and a hyomandibular pouch is absent in osteostracans as that of adult lampreys (Janvier, 1981, 1996b). However, Kuratani and Ahlberg (2018) preferred the first interpretation and the hyomandibular pouch of osteostracans appears to have been developed into a normal full-size gill as in galeaspids. In Shuyu, the canals for the facial nerve and trigeminal nerve have entirely separate courses and there is no evidence for the trigeminal nerve bifurcating into the maxillary and mandibular branches. The condition in Shuyu suggests that the mandibular and the maxillary branch of osteostracans probably always united as the nasopalatine component of trigeminal nerve (V₂,₃, Figure 4D) and passed through the foremost canal together since there are no differentiated upper and lower jaws developed in these stem groups of gnathostomes. Therefore, osteostracans, probably like galeaspids, possess both the velum and a full-size hyomandibular pouch as well.

Various types of water intake device have evolved with the same function as the spiracles of modern rays in jawless fishes, e.g., the prenasal sinus and nasopharyngeal duct in hagfishes, thelodonts and heterostracans, the adorbital openings in amphiaspid heterostracans and pituriaspids (Figure 4H), and the median dorsal opening in galeaspids. These have been considered analogous, but not homologous, to the gnathostome spiracle (Gai et al., 2018). Thus, the hyomandibular pouch probably has been developed as a normal full-size gill pouch (aphetohyoidean condition) in ostracoderms. It has secondarily been lost in cyclostomes because of the evolution of the rasping tongue, which greatly modifies the mandibular arch. In sum, galeaspids provide the first definitive evidence for the presence of a full-size gill pouch between the hyoid and mandibular arches before it was reduced to a spiracle in jawed vertebrates. This observation is not only based on the number and topology of gill arches compared with that of jawed vertebrates, but also on the innervation of the cranial nerve and the impressions of the gill filaments in the first branchial chamber.

While the condition in galeaspids (and probably other ostracoderms) can be described as aphetohyoidean in the sense that a fully developed gill is present between the mandibular and hyoid arches, this does not quite correspond to the “classic” idea of an aphetohyoidean condition in jawed vertebrates, where a fully developed spiracular gill is combined with a mandibular arch developed into jaws. The galeaspid condition would make a perfect evolutionary precursor to the “classic aphetohyoidean” condition, but this would require the retention of the spiracular gill while the mandibular arch was transformed into jaws. Great efforts at finding the aphetohyoidean condition have been made in placoderms, acanthodians and chondrichthyans. Watson (1937) believed that this condition existed in acanthodians (e.g., Acanthodes, Figure 5A), and in placoderms such as arthrodires, petalichthyids and rhenanids, mainly based on the mistaken assumption that their operculum was attached to the mandibular arch, rather than to the hyoid arch as in bony fishes. As the endoskeletal neurocranium and visceral skeleton remain poorly known in early fossil jawed vertebrates, the presence of spiracles has typically been inferred from the spiracular grooves or notches on the cranial bones, as in actinopterygians (Gardiner, 1984; Graham et al., 2014) and sarcopterygians (Jarvik, 1980).

Among acanthodians, the visceral skeleton is only adequately known in Acanthodes from the Lower Permian of Lebach, Germany, a late representative of this derived genus that had to serve as an endoskeletal proxy for all acanthodians. However, the “aphetohyoidean” condition has been doubted by several authors (Holmgren, 1942; Stensiö, 1947; Denison, 1961). Miles (1964, 1965, 1968, 1973) has studied a large number of well-prepared specimens of Acanthodes bronni and Climatius reticulatus, but found no evidence for such an “aphetohyoidean” condition. By contrast, he found the spiracular grooves on the neurocranium in Acanthodes probably containing the spiracular tube. However, as there is no external spiracle opening observed from these tubes, such openings were presumed to be lacking in Acanthodes, rendering the spiracles non-functional. Recently, a very small external spiracular opening was described from the Middle Devonian acanthodiforms Cheiracanthus (Figure 5B) and Mesacanthus in northern Scotland (Burrow et al., 2020). Remarkably, unlike all living jawed vertebrates, these acanthodians prove to have not one but two spiracular pseudobranchs, represented in the fossils by their cartilage supports (Figure 5B). In addition to the posterior, forward-facing, hyoid arch pseudobranch shared with sharks, rays and bony fishes, there is an anterior, backward-facing pseudobranch that must belong to the mandibular arch (Burrow et al., 2020). The spiracle thus contained a complete, albeit miniaturized, gill. Acanthodians are now generally considered to be a group of stem chondrichthyans (Zhu et al., 2013, 2016; Dupret et al., 2014; Burrow et al., 2016; Maisey et al., 2019), which carries the implication that the mandibular-arch spiracular pseudobranch must have been lost independently in osteichthyans and crown-group chondrichthyans. Nevertheless, this is not a fully aphetohyoidean condition, as the gill is miniaturized and restricted to the dorsal space between the palatoquadrate and hyomandibula.

As for the aphetohyoidean condition in placoderms, fossil evidence since the time of Watson (1937) has convincingly demonstrated that the hyoid arch has been strongly modifìed in many placoderms, e.g., a larger hyomandibula in the Rhenanida (Stensiö, 1963), and suggests that they did not have a complete spiracular gill-slit. One of the evidences in support of Watson’s aphetohyoidean theory is that the postsuborbital plate in placoderms has been interpreted as a mandibular operculum guarding a spiracular slit, because of its relations to the quadrate (Watson, 1937). However, the discovery of Entelognathus, and new phylogenetic frameworks, indicate that the posterior mobile submarginal plate in non-Entelognathus placoderms should be considered homologous with the opercular cover of osteichthyans (Zhu et al., 2013; Gai et al., 2017). The submarginal is primitively free in ptyctodontids and primitive arthrodires, but it is enclosed in the cheek and sutured with surrounding bones in advanced arthrodires (Schultze, 1993). A reduced spiracle has been identified in a hyomandibular position in antiarchs (spi, Figures 5G–I; Stensiö, 1947; Johanson, 1998; Arsenault et al., 2004; Young, 2008; Wang and Zhu, 2020) —the most stemward jawed vertebrate group, and in the maxillate placoderm Qilinyu (spi, Figures 5E,F) — representing the most crownward stem gnathostome group (Zhu et al., 2016). Therefore, a restricted spiracular opening may well have existed in all placoderms, but it would be difficult to distinguish it from a gap between dermal elements in most of them (Miyashita, 2016).

The aphetohyoidean theory was later revived for some symmoriiform chondrichthyans, e.g., Cobelodus from Upper Carboniferous black shales of North America (Zangerl and Williams, 1975; Zangerl and Case, 1976), and Ozarcus from the Lower Carboniferous of Arkansas (Pradel et al., 2014). The Symmoriiformes, which were widely distributed from the Late Devonian through to the early Permian, have been regarded as exemplifying early chondrichthyan, and even generalized gnathostome, conditions (Pradel et al., 2011). The hyoid arch of Ozarcus and Cobelodus gives no support to the jaw, and there appears to be a rather wide space between the mandibular and hyoid arches (Figures 5C,D), which suggests that the prehyoidean gill slit had not been reduced to a spiracle in these sharks (Zangerl and Williams, 1975; Zangerl and Case, 1976; Pradel et al., 2014). However, in the earlier symmoriiform shark Ferromirum from the Late Devonian of the Anti-Atlas in Morocco, the three-dimensionally preserved mandibular and hyoid arches leave no room for a prehyoidean gill slit, implying that the apparent gap between mandibular and hyoid arches in Ozarcus and Cobelodus is most probably a post-mortem artifact (Frey et al., 2020). Recent phylogenetic analyses have resolved symmoriiforms as stem holocephalans (Coates et al., 2017, 2018; Frey et al., 2019, 2020), which suggest that they are too derived to represent the primitive jawed vertebrate condition.

In summary, no placoderm or chondrichthyan has yet been shown convincingly to possess a full, aphetohyoidean, spiracular gill slit. Given that osteichthyans also lack such a gill slit (see below), it seems likely that it had been reduced to a dorsally positioned spiracle in the last common ancestor of living jawed vertebrates, i.e., at the gnathostome crown group node. However, the presence of two pseudobranchs in acanthodians, which are believed to be stem chondrichthyans, suggests that full reduction to the modern spiracular condition was a two-step process: an initial size reduction and shift to a dorsal position in the common ancestor of crown gnathostomes was followed, independently in chondrichthyans and osteichthyans, by the loss of the anterior (mandibular arch) pseudobranch. This interpretation implies that two pseudobranchs were also present in placoderms, where no evidence of their attachment survives.

Among bony fishes, the hyomandibular pouch and its derivatives can be observed directly in extant actinopterygians (polypterids, sturgeons and paddlefishes, gar and bowfin, and teleosts) and sarcoptergian fishes (coelacanths and lungfishes). In addition, reasonably robust inferences about its condition in the adult of extinct groups such as tetrapodomorph fishes can be drawn from the morphology of the surrounding skeletal structures of the fossils. Among actinopterygians, an open spiracle broadly similar to that of sharks is found in polypterids, sturgeons and paddlefishes. In Polypterus (Figure 6A), which has a functional lung and is heavily dependent on air-beathing, the large and dorsally positioned spiracle plays an important role in lung ventilation (Graham et al., 2014). Up to 93% of air inhalation occurs through the spiracle. This has important implications for the interpretation of the morphologically similar spiracular region in tetrapodomorph fishes (see below). By contrast, the small spiracles of sturgeons and paddlefishes seem to have no respiratory function (Burggren, 1978; Burggren and Bemis, 1991). In neopterygians (gars, bowfins and teleosts) the spiracles are vestigial or absent, although the pseudobranch often persists inside the gill chamber. Its function is uncertain; it seems not to regulate the hypoxia response (Perry et al., 2011), but may have a role in supplying oxygen to the retina of the eye (Möhlich et al., 2009; Waser, 2011).

Neither of the two living groups of sarcopterygian fishes has a functional spiracle. In the coelacanth Latimeria there is a spiracular chamber (Figure 1H), but this does not open to the outside (Figure 1G) and the pseudobranch is absent (Forey, 1998). In lungfishes, both spiracle and pseudobranch are lost. The spiracular rudiment consists largely of a solid endodermal cellular strand, which never (or only for short time in ontogeny) opens to the outside and the pharyngeal endoderm, respectively (Bartsch, 1994). By contrast, the early bony fishes such as Cheirolepis (stem actinopterygians, Figure 6B; Pearson and Westoll, 1979), Guiyu (stem sarcopterygian, Figure 6D; Zhu et al., 2009), Ligulalepis (stem osteichthyan, Figure 6C; Basden and Young, 2001), Onychodus (Onychodontiformes, Figure 7C; Long, 2001), Porolepis (porolepiformes, Figure 7B) and tetrapodomorph fishes (fossil fishes belonging to the tetrapod stem group) such as Eusthenopteron (Figures 7A, 8A; Jarvik, 1980) and Gogonasus (Figure 6E; Long et al., 2006) always have an open spiracle (spi, Figures 6, 7), judging by the surrounding skeletal anatomy. In Gogonasus the dorsally positioned spiracle is strikingly large (Figure 6E), suggesting an important respiratory role similar to that of Polypterus.

Essentially the same spiracular morphology as in these tetrapodomorphs, but accompanied by a shorter and wider space for the spiracular canal, is found in the elpistostegids Panderichthys (Figure 7D) and Tiktaalik (Figure 6F; Brazeau and Ahlberg, 2006; Downs et al., 2008). Elpistostege is probably similar, though the internal part of the spiracular region is currently unknown (Cloutier et al., 2020). Given the shape of the elpistostegid skull, which suggests a surface-skimming lifestyle, it seems likely that the spiracle was used for air-breathing in a manner similar to Polypterus. This conclusion is further supported by the fact that the external nostril was low on the face, near the upper lip, and would have been submerged in a crocodile-like surface-skimming pose. Interestingly, in both Panderichthys and Tiktaalik the distal part of the hyomandibula has been lost, so that the bone no longer reaches down toward the jaw joint (Brazeau and Ahlberg, 2006; Downs et al., 2008). The corresponding portion of the hyomandibula in Polypterus is the part that lies ventral to the internal opening of the spiracular canal (Datovo and Rizzato, 2018), so it seems highly likely that this truncation of the elpistostegid hyomandibula reflects a change in its relationship to the spiracle, but the exact nature of this change cannot be determined.

Beginning in tetrapods, the hyomandibula ceases to be involved in jaw suspension and instead becomes dedicated eventually to hearing as the stapes (or columella) within the middle ear. Among the earliest Devonian tetrapods, we find the same combination of large, dorsally placed spiracles and small, ventrally facing external nostrils in Parmastega, Ventastega (Figure 6G) and Acanthostega (Figure 8B; Clack, 2003; Ahlberg et al., 2008; Beznosov et al., 2019). This suggests a conserved air-breathing function for the spiracle across the fish-tetrapod transition (Brazeau and Ahlberg, 2006). However, the skeletal morphology of the hyomandibula (henceforth termed “stapes”) and braincase have changed substantially compared to the elpistostegid condition. The most complete example of the earliest tetrapod condition is given by Acanthostega (Figure 6B; Clack, 1998; Clack et al., 2003); the stapes is unknown in Parmastega and Ventastega, though the braincase of Ventastega appears very similar to that of Acanthostega (Ahlberg et al., 2008). In contrast to the fish hyomandibula, which articulates proximally with the lateral commissure (a bony bridge that straddles the jugular vein), the footplate of the tetrapod stapes is lodged in a large opening in the braincase side wall, known as the fenestra vestibuli (fv, Figure 8C2). This puts it in direct contact with the inner ear and suggests that it may have acquired a sound-transducing function even at this very early stage in its evolution. The stapes is also shorter than even the elpistostegid hyomandibula, and is oriented transversely to the long axis of the head.

The stapes of Acanthostega is large, butterfly-shaped and evidently not associated with a tympanum (Figure 8B; Clack, 1998; Clack et al., 2003). Its precise function is difficult to determine, but it was presumably embedded in the posterior wall of the spiracular tract. An essentially similar non-tympanic stapes, usually but not invariably associated with a spiracular notch, is also found in a number of Carboniferous tetrapods including Pederpes, Greererpeton and Pholiderpeton (Figure 6H; Clack et al., 2003). This is thus almost certainly the primitive condition for the tetrapod middle ear. A distinctive variant on the same theme is found in Ichthyostega (Figure 8C1), where the stapes is thin and plate-like, and seems to have formed the floor of a large diverticulum of the spiracular tract (Figure 8C2; Clack et al., 2003). Although it is impossible to fully understand this structure in the absence of the all-important soft tissues, it seems probable that it combined a retained air-breathing function with an aquatic auditory function based on sound waves being picked up by the density contrast between body tissues and the enclosed air space, similar to the use of the swim bladder and Weberian ossicles in ostariophysian teleosts.

Turning to the tetrapod crown group, we find some puzzling patterns that point to multiple origins of the tympanic ear (Figures 8D,E; Clack, 2012). In the temnospondyls, which form the stem group of the extant amphibians, a tympanic ear develops in the same position as the spiracular notch; the transition is documented by the stapes developing a rod-like morphology and the notch acquiring a distinct attachment rim. Unlike Devonian tetrapods, temnospondyls with crocodile-like skulls also have large, dorsally facing external nostrils, which appear to have taken over the air-breathing function previously performed by the spiracles. The earliest well-preserved example of a temnospondyl stapes is that of Dendrerpeton (Figure 7E) from the Late Carboniferous of Canada (Robinson et al., 2005). This type of ear is retained essentially unchanged by modern anurans, whereas urodeles and caecilians have lost the tympanum.

Among crown amniotes, we find that early reptiles such as Captorhinus and early synapsids (stem mammals) such as Ophiacodon lack a tympanic notch and have a large stapes that braces between the quadrate and the braincase. This strongly suggests that the primitive condition for the amniote crown group was non-tympanic, a conclusion that is further supported by the different construction of the middle ears (one ear ossicle or three), and their different position (above the jaw joint or below it), in reptiles (including birds) on the one hand and mammals on the other. The evolution of the mammalian middle ear is well documented in the fossil record (Figure 8E), which shows how it originates at the posteroventral margin of the lower jaw; this is also supported by developmental data (Maier and Ruf, 2016).

Somewhat surprisingly, the stem amniotes Seymouria (Figure 7F) and Diadectes both have tympanic notches that seem to have contained ear drums, and the stapes points toward this notch rather than toward the quadrate; in Diadectes, the tympanum is at least partly ossified (Romer, 1928; Klembara et al., 2020). This type of ear is most similar to that of the temnospondyls, raising the intriguing possibility that a tympanic ear may be primitive for the tetrapod crown group and subsequently lost at the base of the amniote crown group, only to be re-evolved independently in mammals (Figures 8E, 9) on the one hand and reptiles including birds on the other. Critical to answering this question is the phylogenetic position of the anthracosaurs, a Carboniferous to Permian group that includes such genera as Archeria and Pholiderpeton (Figure 6H; Clack, 2012). Anthracosaurs, which are sometimes recovered as stem amniotes and sometimes as stem tetrapods, have a butterfly-shaped stapes and probably had an open spiracle. If they are stem tetrapods, the tympanic ears of stem amniotes and temnospondyls may be homologous, but if anthracosaurs also belong in the amniote stem group these ears are probably convergent. In any case it is clear that tympanic ears evolved at least three times in parallel among the ancestors of the living tetrapods. The persistence of the hyomandibular pouch as an organizing structure in the embryo, coupled with its ability in later development to either break through to the exterior (creating a spiracle), fail to break through but remain separated from the outside world by a thin membrane (creating a tympanum), or recede altogether (creating a non-tympanic ear), provides a mechanistic basis for this remarkable sequence of evolutionary changes.